Papers
Utilizing Marginal Net Utility for Recommendation in E-commerce
Jian Wang, Yi Zhang
Summarizing the Differences in Multilingual News
Xiaojun Wan, Houping Jia
Cross-Language Web Page Classification via Joint Nonnegative Matrix Tri-factorization Based Dyadic K
Wang Hua, Heng Huang
Incremental Diversification for Very Large Sets: a Streaming-based Approach
Enrico Minack, Wolf Siberski, Wolfgang Nejdl
Multifaceted Toponym Recognition for Streaming News
Michael Lieberman, Hanan Samet
Collaborative Competitive Filtering: Learning Recommender Using Context of User Choices
Shuang-Hong Yang, Bo Long, Alex Smola, Hongyuan Zha, Zhaohui Zheng
Who Should Share What? Item-level Social Influence Prediction for Users and Posts Ranking
Cui Peng, Fei Wang, Shao-Wei Liu, Ming-Dong Ou, Shi-Qiang Yang
Identifying Points of Interest by Self-Tuning Clustering
YiYang Yang, Zhiguo Gong, Leong Hou U
PICASSO – To Sing you must Close Your Eyes and Draw
Aleksandar Stupar, Sebastian Michel
Collective Entity Linking in Web Text: A Graph-Based Method
Xianpei Han, Le Sun
Toward Social Context Summarization For Web Documents
Zi Yang, cai keke, Jie Tang, Li Zhang, Zhong Su, Juanzi Li
CLR: A Collaborative Location Recommendation Framework based on Co-Clustering
Kenneth Wai-Ting Leung, Wang-Chien Lee, Dik Lun Lee
A Boosting Approach to Improving Pseudo-Relevance Feedback
Yuanhua Lv, Wan Chen, ChengXiang Zhai
Detecting Outlier Sections in US Congressional Legislation
Elif Aktolga, Irene Ros, Yannick Assogba
Out of sight, not out of mind: On the effect of social and physical detachment on information need
Elad Yom-Tov, Fernando Diaz
Synthesizing High Utility Suggestions for Rare Web Search Queries
Alpa Jain, Umut Ozertem, Emre Velipasaoglu
Measuring Improvement in User Search Performance Resulting From Optimal Search Tips
Neema Moraveji, Daniel Russell, Jacob Bien, David Mease
ILDA: Interdependent LDA Model for Learning Latent Aspects and their Ratings from Online Product Rev
Samaneh Moghaddam, Martin Ester
Evolutionary Timeline Summarization: a Balanced Optimization Framework via Iterative Substitution
Rui Yan, Xiaojun Wan, Jahna Otterbacher, Liang Kong, Yan Zhang, Xiaoming Li
Efficient Manifold Ranking for Image Retrieval
Bin Xu, Jiajun Bu, Chun Chen, Zhanying He, Deng Cai
The Economics in Interactive Information Retrieval
Leif Azzopardi
Cluster-based fusion of retrieved lists
Anna  Khudyak Kozorovitsky, Oren Kurland
Active Learning to Maximize Accuracy vs. Effort in Interactive Information Retrieval
Aibo Tian, Matthew Lease
Document Clustering with Universum
Dan Zhang, Jingdong Wang, Luo Si
Composite Hashing with Multiple Information Sources
Dan Zhang, Fei Wang, Luo Si
Automatic Boolean Query Suggestion for Professional Search
Youngho Kim, Jangwon Seo, Bruce Croft
Inverted Indexes for Phrases and Strings
Manish Patil, Sharma V. Thankachan, Rahul Shah, Wing-Kai Hon, Jeffrey Vitter, Sabrina Chandrasekaran
Query by Document via a Decomposition-Based Two-Level Retrieval Approach
Linkai Weng
UPS: Efficient Privacy Protection in Personalized Web Search
He Bai, Lidan Shou, Ke Chen, Gang Chen, Yunjun Gao
Learning Search Tasks in Queries and Web Pages via Graph Regularization
Ming Ji, Jun  Yan, Siyu Gu, Jiawei Han, Xiaofei He
Multimedia Answering: Enriching Text QA with Media Information
Liqiang Nie, Meng Wang, Zha Zhengjun, Li Guangda, Tat Seng Chua
Evaluating Diversified Search Results Using Per-intent Graded Relevance
Tetsuya Sakai, Ruihua Song
Functional Matrix Factorizations for Cold-Start Collaborative Filtering
Ke Zhou, Shuang-Hong Yang, Hongyuan Zha
Mining Topics on Participations for Community Discovery
Guoqing Zheng, Jinwen Guo, Lichun Yang, Shengliang Xu, Shenghua Bao, Zhong Su, Dingyi Han, Yong Yu
Integrating Hierarchical Feature Selection and Classifier Training for Multi-Label Image Annotation
Cheng Jin, Xiangyang Xue
Understanding Re-finding Behaviour in Naturalistic Email Interaction Logs
David Elsweiler, Morgan  Harvey, Martin  Hacker
Bagging Gradient-Boosted Trees for High Precision, Low Variance Ranking Models
Yasser Ganjisaffar, Rich Caruana, Cristina Lopes
Query Suggestions in the Absence of Query Logs
Sumit Bhatia, Debapriyo  Majumdar, Prasenjit Mitra
Estimation Methods for Ranking Recent Information
Miles Efron, Gene Golovchinsky
Unsupervised Query Segmentation Using Clickthrough for Information Retrieval
Yanen Li, Bo-June Hsu, ChengXiang Zhai, Kuansan Wang
Pseudo Test Collections for Learning Web Search Ranking Functions
Nima Asadi, Donald Metzler, Tamer Elsayed, Jimmy Lin
Predicting Web Searcher Satisfaction with Existing Community-based Answers
Qiaoling Liu, Eugene Agichtein, Gideon Dror, Evgeniy Gabrilovich, Yoelle Maarek, Dan Pelleg, Idan Szpektor
Social Annotation in Query Expansion a Machine Learning Approach
Yuan Lin, Song Jin, Hongfei Lin, Yunlong Ma, Kan Xu
Probabilistic Factor Models for Web Site Recommendation
Hao Ma, Chao Liu, Irwin King, Michael R. Lyu
Quaternary Semantic Analysis: Providing Recommendations based on Explicit and Implicit Feedbacks
Chen wei, Hsu Wynne, Mong Li Lee
Mining Tags Using Social Endorsement Networks
Theodoros Lappas, Kunal Punera, Tamas Sarlos
Why Searchers Switch: Understanding and Predicting Engine Switching Rationales
Qi Guo, Ryen White, Yunqiao Zhang, Blake Anderson, Susan Dumais
Exploiting Geographical Influence for Collaborative Point-of-Interests Recommendation
Mao Ye, Peifeng Yin, Wang-Chien Lee, Dik Lun Lee
Clickthrough-Based Latent Semantic Models for Web Search
Jianfeng Gao, Kristina Toutanova, Wen-tau Yih
Indexing Strategies for Graceful Degradation of Search Quality
Shuai Ding, Sreenivas Gollapudi, Samuel Ieong, Krishnaram Kenthapadi, Alexandros Ntoulas
Enriching Document Representation via Translation for Improved Monolingual Information Retrieval
Seung-Hoon Na, Hwee Tou Ng
Regularized Latent Semantic Indexing
Quan Wang, Jun Xu, Hang Li, Nick Craswell
People Searching for People: Analysis of a People Search Engine Log
Wouter Weerkamp, Richard Berendsen, Bogomil Kovachev, Edgar Meij, Krisztian Balog, Maarten de Rijke
An Event-centric Model for Multilingual Document Similarity
Jannik Strötgen, Conny Junghans, Michael Gertz
From One Tree to a Forest: a Unified Solution for Structured Web Data Extraction
Qiang Hao, Rui Cai, Lei Zhang
A Site Oriented Method For Segmenting Web Pages
David Fernandes de Oliveira, Edleno Moura, Altigran da Silva, Berthier Ribeiro-Neto, Edisson Braga
Learning to Rank for Freshness and Relevance
Na Dai, Milad Shokouhi, Brian Davison
A Novel Corpus-Based Stemming Algorithm  using Co-occurrence Statistics
Jiaul Paik, Dipasree  Pal, Swapan  Parui
DOM Based Content Extraction via Text Density
Fei Sun, Dandan Song, Lejian Liao
Scalable multi-dimensional user intent identification using tree structured distributions
Vinay Jethava, Liliana Calderon-Benavides, Chiranjib Bhattacharyya, Devdatt Dubhashi, Ricardo Baeza-Yates
Handling Data Sparsity in Collaborative Filtering using Emotion and Semantic Based Features
Yashar Moshfeghi, Benjamin Piwowarski, Joemon M. Jose
Improved video categorization from text metadata and user comments
Katja Filippova, Keith Hall
Distributed, Private, and Anonymous Search Logs
Henry Feild, James Allan, Joshua Glatt
Evaluating Multi-Query Sessions
Evangelos Kanoulas, Ben Carterette, Paul Clough, Mark Sanderson
Parameterized Concept Weighting in Verbose Queries
Michael Bendersky, Donald Metzler, Bruce Croft
Mining Weakly Labeled Web Facial Images for Search-based Face Annotation
Steven Hoi, dayong wang, Ying He
Fast Context-aware Recommendations with Factorization Machines
Steffen Rendle, Zeno Gantner, Christoph Freudenthaler, Lars Schmidt-Thieme
No Free Lunch: Brute Force vs. Locality-Sensitive Hashing for Cross-lingual Pairwise Similarity
Ferhan Ture, Tamer Elsayed, Jimmy Lin
Dynamics of Focus-of-Attention in Diagnostic Search
Marc-Allen Cartright, Ryen White, Eric Horvitz
CRTER: Using Cross Terms to Enhance Probabilistic Information Retrieval
Jiashu Zhao, Jimmy Huang, Ben He
User Behavior in Zero-Recall eCommerce Queries
Gyanit Singh, Nish Parikh, Neel Sundaresan
Faster Temporal Range Queries over Versioned Text
Jinru He, Torsten Suel
Hypergeometric Language Models for Republished Article Finding
Manos Tsagkias, Maarten de Rijke, Wouter Weerkamp
SCENE : A Scalable Two-Stage Personalized News Recommendation System
Lei Li, Dingding Wang, Tao Li
Crowdsourcing for Book Search Evaluation: Impact of Quality on Comparative System Ranking
Gabriella Kazai, Jaap Kamps, Marijn Koolen, Natasa Milic-Frayling
Post-Ranking Query Suggestion by Diversifying Search Results
Yang Song, Dengyong Zhou, Li-wei He
Authorship Classification: A Discriminative Syntactic Tree Mining Approach
Sangkyum Kim, Hyungsul Kim, Tim Weninger, Jiawei Han, Hyun Duk Kim
Learning Online Discussion Structures by Conditional Random Fields
Wang Hongning, Chi Wang, ChengXiang Zhai, Jiawei Han
Filtering Semi-Structured Documents Based on Faceted Feedback
Lanbo Zhang, Yi Zhang
A Cascade Ranking Model for Efficient Ranked Retrieval
Lidan Wang, Jimmy Lin, Donald Metzler
Ranking Related News Predictions
Nattiya Kanhabua, Roi Blanco, Michael Matthews
Modeling and Analysis of Cross-Session Search Tasks
Alexander Kotov, Paul Bennett, Ryen White, Susan Dumais, Jaime Teevan
Repeatable and Reliable Search System Evaluation using Crowd-Sourcing
Roi Blanco, Harry Halpin, Daniel  Herzig, Peter Mika, Jeffrey Pound, Henry Thompson, Thanh D. Tran
Enhanced Results for Web Search
Kevin Haas, Peter Mika, Paul Tarjan, Roi Blanco
Efficiently Collecting Relevance Information from Clickthroughs for Web Retrieval System Evaluation
Jing He, Xin Zhao, Baihan Shu, Xiaoming Li, Hongfei Yan
System Effectiveness, User Models, and User Utility:  A General Framework for Investigation
Ben Carterette
Seeding Simulated Queries with User-study Data for Personal Search Evaluation
David Elsweiler, David Losada, Jose Carlos  Toucedo, Ronald T. Fernández
Associative Tag Recommendation Exploiting Multiple Textual Features
Fabiano Belém, Eder Martins, Tatiana Pontes, Jussara Almeida, Marcos Goncalves
Recommending Ephemeral Items at Web Scale
Ye Chen, John Canny
Improving Local Search Ranking through External Logs
Klaus Berberich, König Arnd, Dimitrios  Lymberopoulos, Peixiang Zhao
Intent-Aware Search Result Diversification
Rodrygo Santos, Craig Macdonald, Iadh Ounis
Inferring and Using Location Metadata to Personalize Web Search
Paul Bennett, Filip Radlinski, Ryen White, Emine Yilmaz
Evaluating the Synergic Effect of Collaboration in Information Seeking
Chirag Shah, Roberto González-Ibáñez
Temporal Index Sharding for Space-Time Efficiency in Archive Search
Avishek Anand, Srikanta Bedathur, Ralf Schenkel, Klaus Berberich
Energy-Price-Driven Query Processing in Multi-center Web Search Engines
Enver Kayaaslan, B. Barla Cambazoglu, Roi Blanco, Cevdet Aykanat
Learning Sentiment Streams with Training Expansion and Demand-Driven Partitioning
Ismael Santana Silva, Janaína Gomide, Adriano Veloso, Renato Ferreira, Wagner Meira Jr.
ViewSer: Enabling Large-Scale Remote User Studies of Web Search Examination and Interaction
Dmitry Lagun, Eugene Agichtein
Posting List Intersection on Multicore Architectures
Shirish Tatikonda, B. Barla Cambazoglu, Flavio Junqueira
Timestamp-based Result Cache Invalidation Mechanisms for Web Search Enginess
Adiye Alici, Ismail Altingovde, Rifat Ozcan, B. Barla Cambazoglu, Ozgur Ulusoy
Enhancing Multi-Label Music Genre Classification Through Ensemble Techniques
Chris Sanden, John Zhang
Learning Relevance in a Heterogeneous Social Network and Its Application in Online Targeting
Chi Wang, Rajat Raina, David Fong, Ding Zhou, Jiawei Han, Greg Badros
Find It If You Can: Modeling and Predicting Different Types of Web Search Success with Behavior Data
Mikhail Ageev, Qi Guo, Dmitry Lagun, Eugene Agichtein
Faster Top-k Document Retrieval Using Block-Max Indexes
Ding Shuai, Torsten Suel
Enhancing Ad-hoc Relevance Weighting Using Probability Density Estimation
Xiaofeng Zhou, Jimmy Huang, Ben He
On Theme Location Discovery for Travelogue Services
Mao Ye, Rong Xiao, Wang-Chien Lee, Xing Xie
Quantifying test collection quality based on the consistency of relevance judgements
Falk Scholer, Andrew Turpin, Mark Sanderson
Competition-based User Expertise Level Estimation
Jing Liu, Young-In Song, Chin-Yew Lin
Relevant Knowledge Helps in Choosing Right Teacher: Active Query Selection for Ranking Adaptation
Peng Cai, Wei Gao, Aoying Zhou, Kam-Fai Wong

SIGIR

Sparse is very hot in compressive sensing research.  When I just read a paper,  an idea got to my mind. Why not register a related domain for it? I love this idea, and search for sparselab.com, amazing it is available!

It cost me arround 9 dollars to get it at namecheap instead of godaddy, because namecheap is cheaper:)

Compressive Sensing · sparse

Interdisciplinary Workshop
SPARSITY
AND
MODERN MATHEMATICAL METHODS
FOR
HIGH DIMENSIONAL DATA
April 6–April 10, 2010
Vrije Universiteit Brussel
Version: April 3, 2010
Organization
•  Ingrid Daubechies (Princeton University)
•  Christine De Mol (Universit´e Libre de Bruxelles)
•  Ignace Loris (Vrije Universiteit Brussel)
•  Benoit Macq (Universit´e Catholique de Louvain)
•  Aleksandra Pizurica (Universiteit Gent)
•  Philippe Cara (Vrije Universiteit Brussel)
•  Ann Dooms (Vrije Universiteit Brussel)
•  Caroline Verhoeven (Vrije Universiteit Brussel)
Conference website: http://www.sparsity.be/
Sponsors
•  FWO-Vlaanderen

http://www.fwo.be/

•  International Solvay Institutes

http://www.solvayinstitutes.be/

•  FWO SRN Advanced Numerical Methods for Mathematical Modelling

http://www.cs.kuleuven.be/˜ade/WWW/WOG/

•  FWO SRN Audio-Visual Systems

http://www.avs.vub.ac.be/

•  Francqui Foundation

http://www.francquifoundation.be/

•  Vrije Universiteit Brussel

http://www.www.vub.ac.be/

•  Interdisciplinary Institute for Broadband Technology

http://www.ibbt.be/

2
Contents
Organization 2
Sponsors 2
1    Schedule 4
2    List of presentations 5
2.1    Tuesday, April 6th .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .      5
2.2    Wednesday April 7th   .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .      6
2.3    Thursday April 8th   .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .      8
2.4    Friday April 9th  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .      9
2.5    Saturday April 10th  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    10
3    List of abstracts 11
4    List of participants 37
5    Practical information 42
5.1    Lunch  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    42
6    Getting around 42
6.1    Campus map    .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    43
6.2    By car  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    44
6.3    By public transport   .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    45
6.4    Near hotel “Adagio” .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    46
7    Personal notes 47
Workshop announcement 48
3
1    Schedule
Tuesday Wednesday          Thursday Friday Saturday
9.00–9.50         Registration         Vandergheynst         Portilla Kutyniok        Van De Ville
9.45: Opening
9.50–10.40         Daubechies Kutyniok         Van De Ville     Vandergheynst          Stork
10.40–11.10 (Group photo) Break
11.10–12.00 Portilla Johnson Stork Portilla Daubechies
End
12.00–13.30 Lunch
13.30–14.00 Jansen Fornasier Jacques Rauhut
14.00–14.30 Zhariy De Lathauwer         Krahmer Lee
14.30–15.00    Sankaranarayanan       Van Huffel Dyer Anzengruber
15.00–15.30 Break
15.30–16.00 Mary Huybrechs       Temerinac-Ott       Signoretto
16.00–16.30        Bourguignon Damerval Goossens Gillis
16.30–17.00 Richards Aelterman Tenorth Bleyer
17.00–17.30 Platisa Van de Plas          Haskovec Alzate
17.30–18.00          Discussion Discussion          Discussion          Discussion
18:00–19:00 Poster
Session
19.00–. . . . . . Conference
Dinner
2    List of presentations
2.1    Tuesday, April 6th
REGISTRATION   . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 09:00–9:45
OPENING  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 09:45–9:50
INGRID DAUBECHIES (Princeton University) . . . . . . . . . . . . . . . . . . . . . . . . 09:50–10:40
Introduction to the world of wavelets, curvelets (and other –lets)
COFFEE BREAK  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10:40–11:10
JAVIER PORTILLA (Instituto de Optica, CSIC) . . . . . . . . . . . . . . . . . . . . . . . 11:10–12:00
Sparse Approximation: A general discussion and a simple algorithm
LUNCH   . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12:00–13:30
MAARTEN JANSEN (K.U.Leuven) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13:30–14:00
Minimum loss penalties for iterative soft- and hard-thresholding
MARIYA ZHARIY (Johannes Kepler University) . . . . . . . . . . . . . . . . . . . . . . 14:00–14:30
Truncated Soft Shrinkage Iteration with Discrepancy Based Stopping Rule. Applica-
tion to Inverse Problems with Data Noise.
ASWIN SANKARANARAYANAN (Rice University) . . . . . . . . . . . . . . . . . . . . 14:30–15:00
Compressive acquisition of dynamic scenes
COFFEE BREAK  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15:00–15:30
DAVID MARY (University of Nice). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15:30–16:00
Properties of thresholding functions for different sparsity-based denoising scenarios
S ´EBASTIEN BOURGUIGNON (Universit´e de Nice) . . . . . . . . . . . . . . . . . . . 16:00–16h30
Sparsity-based denoising and source detection in astronomical hyperspectral data
with non iid noise
JOSEPH RICHARDS (Carnegie Mellon University) . . . . . . . . . . . . . . . . . . . . 16:30–17:00
Sparse Prototyping for Astrophysical Spectra
LJILJANA PLATISA (UGent) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17:00–17:30
Image Blur Estimation Based on the Average Cone of Ratio in the Wavelet Domain
5
2.2    Wednesday April 7th
PIERRE VANDERGHEYNST (EPFL) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 09:00–09:50
Wavelets on graphs via spectral theory
GITTA KUTYNIOK (Universit¨at Osnabr¨uck) . . . . . . . . . . . . . . . . . . . . . . . . . .09:50–10:40
Beyond Wavelets: Compactly Supported Shearlets
COFFEE BREAK  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10:40–11:10
DON JOHNSON (Rice University) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11:10–12:00
Signal Processing and Analyzing Works of Art
LUNCH   . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12:00–13:30
MASSIMO FORNASIER (Austrian Academy of Sciences) . . . . . . . . . . . . . . 13:30–14:00
The mathematics of swarming and classical dimensionality reduction principles from
kinetic theory
LIEVEN DE LATHAUWER (K.U.Leuven) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14:00–14:30
Exploiting the Intrinsic Low Dimensionality by Means of Tensor Methods
SABINE VAN HUFFEL (Katholieke Universiteit Leuven) . . . . . . . . . . . . . . 14:30–15:00
Tensor-based biosignal processing
COFFEE BREAK  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15:00–15:30
DAAN HUYBRECHS (KULeuven) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15:30–16:00
Fourier series on triangles and higher-dimensional simplices
CHRISTOPHE DAMERVAL (Joint Research Center of the Eur. Comm.) . . 16:00–16:30
Extraction of regularity features on non-regular grids
JAN AELTERMAN (Ghent University) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16:30–17:00
A Bregman iteration Algorithm for Shearlet-regularized Compressed Sensing in MRI
RAF VAN DE PLAS (K.U.Leuven) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17:00–17:30
Wavelet approaches to enable the exploration of organic tissue via multivariate anal-
ysis of mass spectral imaging data
POSTER SESSION  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18:00–19:00
SAMIR KUMAR BHOWMIK (University of Amsterdam) . . . . . . . . . . . . . . . . . . . (poster)
Windowed Fourier frames applied to partial differential equations
NABIOLLAH G. CHEGINI (Korteweg-de Vries Inst. for Mathematics) . . . . . . .(poster)
An Adaptive Tensor Product Wavelet Scheme for solving PDE’s
6
PHILIPPE DREESEN (KULeuven) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (poster)
Zero-norm minimization as a polynomial optimization problem
MARKO FILIPOVIC (Rudjer Boskovic Institute) . . . . . . . . . . . . . . . . . . . . . . . . . . (poster)
Unsupervised segmentation of multispectral images by sparse component analysis
SILVIA GANDY (Tokyo Institute of Technology) . . . . . . . . . . . . . . . . . . . . . . . . . . (poster)
A study of multiple measurement approaches for low-rank matrix recovery
MARIAN-DANIEL IORDACHE (University of Extremadura) . . . . . . . . . . . . . . . . (poster)
Spectral Libraries for Sparse Hyperspectral Unmixing
JONAS OFFTERMATT (University of Stuttgart) . . . . . . . . . . . . . . . . . . . . . . . . . . . . (poster)
An adaptive algorithm for providing sparse solutions in an application of systems
biology
ANDREA RUEDA-OLARTE/EDUARDO ROMERO (U. N. de Colombia) . . . . . . (poster)
Super-Resolution of Brain MR Images based on Sparse Representations
KHALID SABRI (LATT-CNRS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (poster)
Efficient implementation of greedy algorithms for sparse images deconvolution
GEORGE TZAGKARAKIS (University of Crete & FORTH) . . . . . . . . . . . . . . . . . (poster)
Bayesian Compressed Sensing using Alpha-Stable Distributions
NICO VERBEECK (K.U. Leuven) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .(poster)
Improved Wavelet Analysis of Mass Spectral Imaging Data for Feature Selection and
Data Compression through Incorporation of Spatial Information
SERGEY VORONIN (Princeton University) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (poster)
ℓ1-regularization for applications
7
2.3    Thursday April 8th
JAVIER PORTILLA (Instituto de Optica, CSIC) . . . . . . . . . . . . . . . . . . . . . . . 09:00–09:50
Efficient ℓ0-based sparse approximation using Parseval frames
DIMITRI VAN DE VILLE (EPFL and University of Geneva) . . . . . . . . . . . 09:50–10:40
Surfing the Brain: Wavelets and Sparsity for Functional Brain Imaging
COFFEE BREAK  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10:40–11:10
DAVID STORK (Ricoh Innovations) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11:10–12:00
Computer graphics in the history and interpretation of art: Computer science, optics
and art history confront a bold theory
LUNCH   . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12:00–13:30
LAURENT JACQUES (University of Louvain) . . . . . . . . . . . . . . . . . . . . . . . . . 13:30–14:00
Dequantizing Non-Uniformly Quantized Compressed Sensing: Weight and See!
FELIX KRAHMER (Universit¨at Bonn) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14:00–14:30
Lower Bounds for the Error Decay in One-Bit Quantization
EVA DYER (Rice University) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14:30–15:00
Sparse Coding in Modular Networks
COFFEE BREAK  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15:00–15:30
MAJA TEMERINAC-OTT (University of Freiburg) . . . . . . . . . . . . . . . . . . . . 15:30–16:00
Multichannel Image Restoration Based on Optimization of the Structural Similarity
Index
BART GOOSSENS (Ghent University) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16:00–16:30
Solving Various Problems In Image Restoration Through Shearlet-based Regulariza-
tion
STEFANIE TENORTH (University of Duisburg-Essen) . . . . . . . . . . . . . . . . . 16:30–17:00
A Hybrid Algorithm for Image Approximation Based on the EPWT
JAN HASKOVEC (Austrian Academy of Sciences) . . . . . . . . . . . . . . . . . . . . 17:00–17:30
Mathematical methods for spectral image reconstruction
8
2.4    Friday April 9th
GITTA KUTYNIOK (Universit¨at Osnabr¨uck) . . . . . . . . . . . . . . . . . . . . . . . . . .09:00–09:50
Beyond Sparsity: Clustered Sparsity and Data Separation
PIERRE VANDERGHEYNST (EPFL) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 09:50–10:40
Spread spectrum imaging techniques in MRI and Radio-interferometry: experimental
promises
COFFEE BREAK  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10:40–11:10
JAVIER PORTILLA (Instituto de Optica, CSIC) . . . . . . . . . . . . . . . . . . . . . . . 11:10–12:00
From approximation to estimation. Some image processing examples.
LUNCH   . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12:00–13:30
HOLGER RAUHUT (University of Bonn) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13:30–14:00
Sparse Legendre expansions via ℓ1-minimization
WEN-SHIN LEE (University of Antwerp) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14:00–14:30
Reconstructing a Sparse Trigonometric Polynomial
STEPHAN W. ANZENGRUBER (RICAM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14:30–15:00
Morozov’s Discrepancy Principle for Tikhonov-type regularization
COFFEE BREAK  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15:00–15:30
MARCO SIGNORETTO (K.U. Leuven) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15:30–16:00
High Dimensional Sparse Estimation with Multiple Graphs
NICOLAS GILLIS (Universit´e catholique de Louvain) . . . . . . . . . . . . . . . . . 16:00–16:30
Dimensionality reduction, classification, and spectral mixture analysis using nonneg-
ative underapproximation
ISMAEL RODRIGO BLEYER (Johannes Kepler University) . . . . . . . . . . . . 16:30–17:00
Regularization of linear integral equations with noisy data and noisy operator
CARLOS ALZATE (Katholieke Universiteit Leuven) . . . . . . . . . . . . . . . . . . 17:00–17:30
Highly Sparse Model Representations for Kernel Spectral Clustering
9
2.5    Saturday April 10th
DIMITRI VAN DE VILLE (EPFL and University of Geneva) . . . . . . . . . . . 09:00–09:50
Steerable Wavelet Pyramids and Reconstruction from a Compact Multiscale Primal
Sketch
DAVID STORK (Ricoh Innovations) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 09:50–10:40
Did the great masters “cheat” using optics? Computer science, optics and art history
confront a bold theory
COFFEE BREAK  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10:40–11:10
INGRID DAUBECHIES (Princeton University) . . . . . . . . . . . . . . . . . . . . . . . . 11:10–12:00
Sparse expansions: what have we learned and where are we going?
WORKSHOP CLOSING  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12:00–12:05
10
3    List of abstracts
Jan Aelterman (Ghent University)
A Bregman iteration Algorithm for Shearlet-regularized Compressed Sensing in MRI
Recently, it has been shown that MRI acquisition can be improved a lot using Com-
pressive Sensing (CS) techniques. In our work we focus on reconstructing sub-Nyquist
sampled MRI data, which we regularize using the shearlet transform.  The shearlet
transform is credited as providing an optimally sparse frame for representing smooth
image regions delineated by edges. Hence, it is a good model for MRI images. The
resulting basis pursuit (BP) CS formulation is solved using split Bregman iteration,
which splits the BP problem into several easier subproblems. The resulting algorithm
allows an exact, parameter-free solution to the constrained BP problem.  The results
show that the algorithm is able to perform any MRI reconstruction task (sub-Nyquist
sampled data or not) and even perform image fusion and resolution enhancement.
Carlos Alzate (Katholieke Universiteit Leuven)
Highly Sparse Model Representations for Kernel Spectral Clustering
Classical spectral clustering methods are relaxations of NP-hard graph partitioning
problems. These relaxations lead to eigenvalue problems of a graph Laplacian matrix.
The relaxed solutions correspond to particular eigenvectors which contain the group-
ing information.  A different formulation to spectral clustering has been introduced
lately as a primal-dual optimization setting allowing natural out-of-sample extensions
and model selection. The formulation in the dual corresponds to an eigenvalue prob-
lem and can be interpreted as a weighted version of kernel PCA for a particular choice
of weights. The clustering model can be extended to new points via projections onto
the eigenvectors which is important for obtaining a good generalization performance.
The eigenvectors and the projections display a special structure when the clusters are
well-formed and the kernel function and its parameters have been chosen appropri-
ately. Data points in the same cluster are collinear in the space spanned by the projec-
tions. However, these projections are expressed in terms of non-sparse kernel expan-
sions where every training data point contributes. A highly sparse method for spectral
clustering is presented in this work.  This sparse scheme works in two levels: first a
representative set of N training points is obtained by maximizing the quadratic Renyi
entropy on the full dataset of size N full with N ≪ N full . Second, the projections are
approximated using a reduced set method of R data points with R ≪ N. The choice
of these R data points is based on particular positions on the lines that represent the
clusters. The proposed approach leads to highly sparse clustering models with predic-
tive capabilities. Experimental results with toy data and images show the effectiveness
of our method.
Joint work with Johan A. K. Suykens.
11
Stephan W. Anzengruber (Radon Institute for Computational and Applied
Mathematics)
Morozov’s Discrepancy Principle for Tikhonov-type regularization
In this talk we will be concerned with Tikhonov regularization using general convex
penalty terms (including the sparsity promoting choice Ψ p = k.k p, for 1 ≤ p ≤ 2)
for nonlinear inverse problems and choose Morozov’s discrepancy principle as the
parameter choice rule.  We will discuss when the discrepancy principle is applicable
and show that when using the penalty terms Ψ p, the regularized solutions converge to
a Ψ p-minimizing solution even with respect to Ψ p. A new condition will be presented
which ensures that the regularization parameter chosen according to the discrepancy
principle approaches zero as the noise in the data goes to zero. Finally, we will discuss
convergence rates in the Bregman distance under standard nonlinearity and source
conditions and present numerical results reconstructing a sparse solution with respect
to a wavelet basis for the autoconvolution operator.
Samir Kumar Bhowmik (University of Amsterdam)
Windowed Fourier frames applied to partial differential equations
We investigate the application of windowed Fourier frames (WFFs) to the numerical
solution of partial differential equations, focussing on elliptic equations.  The action
of a partial differential operator (PDO) on a windowed plane wave is close to a multi-
plication, where the multiplication factor is given by the symbol of the PDO evaluated
at the wave number and central position of the windowed plane wave.  This can be
exploited in a preconditioning method for use in iterative inversion. We also study the
approximation of operator matrix elements in a windowed Fourier frame.
Ismael Rodrigo Bleyer (Doktoratskolleg ”Computational Mathematics”)
Regularization of linear integral equations with noisy data and noisy operator
Regularization methods for linear ill-posed problems Kf = g have been extensively
investigated when there exists noisy measurements g
δ
for the true data g.  However,
often also the operator is not known exactly. A common way to solve this problem is
to use the regularized total least squares method. In our approach, we consider linear
integral equations where we assume that both the kernel and the data are contami-
nated with noise, and use Tikhonov regularization for a stable reconstruction of both
the kernel and the solution.  Finally, we discuss computational aspects and develop
an iterative shrinkage-thresholding algorithm for the reconstruction, and provide first
numerical results.
S´ebastien Bourguignon (Universit´e de Nice)
Sparsity-based denoising and source detection in astronomical hyperspectral data
with non iid noise
We consider the denoising of hyperspectral data for the astronomical MUSE (Multi-
Unit Spectroscopic Explorer) integral field  spectrograph,  which will provide data
12
cubes with 300×300 pixels and up to 4000 spectral channels.  MUSE observations
of the distant universe will be collected in a very noisy environment, mainly caused
by the powerful parasite emission in the atmosphere at specific wavelengths. Conse-
quently, high spectral variations occur in the noise level. Moreover, taken spectrally,
data also suffer from spreading by MUSE’ Line Spread Function (LSF), which is not
constant in the considered spectral range.  Hence, the spreading is not a convolution
operation.
In such a framework, we consider the problem of data denoising based on sparsity
constraints imposed on the spectra to be reconstructed. At each pixel, the spectrum is
supposed to have a sparse representation in a union of bases, e.g. the canonical one (for
emission and absorption lines) and another base such as Discrete Cosine or Wavelet
Transform, in which most of the continuous part of the spectrum is supposed to be
well approximated by a few coefficients. We classically consider the minimization of
a quadratic data-fit functional, penalised by ℓ1-norm constraints.  The originality of
this contribution is twofold.  First, the functional takes into account the noise spec-
tral distribution, both in the data attachment term and in the penalisation term, where
hyperparameter values are linked to the noise spectral distribution by statistical argu-
ments. Then, algorithmic issues are investigated. In particular, because of noise non-
iidness and of the variability of the LSF, matrix-vector products cannot be computed
by fast transform algorithms, making a large class of recently developed methods in-
efficient in our case. We propose a convergent strategy that mixes coordinate descent
steps with Iterative Re-weighted Least-Squares, which reveal complementary proper-
ties for retrieving sparse solutions. Results of such ℓ1 minimization are compared with
greedy algorithms for approximate ℓ0 minimization, which are also adapted to the non
iid framework.
Nabiollah Godarzvand Chegini         (Korteweg-de Vries institute for mathematics)
An Adaptive Tensor Product Wavelet Scheme for solving PDE’s
Adaptive wavelet schemes solve PDE’s with the best possible convergence rate in lin-
ear complexity. Moreover, when tensor product wavelets are applied, this rate can be
shown to be independent of the space dimension.  We construct a univariate wavelet
basis such that any constant coefficient second order PDE with respect to the tensor
product basis yields an infinite stiffness matrix that is sparse.  This drastically sim-
plifies the implementation and improves the quantitative properties of the adaptive
wavelet scheme. We illustrate our findings with numerical results.
Christophe Damerval (Joint Research Center of the European Commission)
Extraction of regularity features on non-regular grids
The Lipschitz regularity is a value α ∈ R that can measure the local regularity of
a function.  Various works studied this feature in several domains: signal and image
processing, computer vision. Its robustness under geometric and photometric transfor-
mations emphasizes its relevance for the applications (extraction of invariant features).
Here we address the problem of the estimation of the Lipschitz regularity. While cur-
13
rent methods assume that data are given on regular grids, we propose here to deal with
non-regular ones. Multiscale methods provide fast numerical computations, allowing
to deal with irregularly sampled time series (1D signals), and 2D surfaces on irregular
meshes.
Keywords: Lipschitz regularity, wavelets, lifting scheme, irregular subdivisions.
Ingrid Daubechies (Princeton University)
Introduction to the world of wavelets, curvelets (and other –lets)
Wavelets are 25 to 30 years old now.   With hindsight, we can see that they were
succesful for images because they led to a sparser representation than other methods
existing at the time, and to nonlinear approximation.  They were also only the first
arrival of a larger family of -lets: curvelets, shearlets, . . . This talk will introduce the
larger framework in which wavelets and more recent developments fit.
Ingrid Daubechies (Princeton University)
Sparse expansions: what have we learned and where are we going?
This talk aims to summarize the different aspects of sparse decompositions presented
during the workshop, as well as other recent developments. At the end, we can try to
guess what the (near) future will bring . . .
Lieven De Lathauwer (K.U.Leuven)
Exploiting the Intrinsic Low Dimensionality by Means of Tensor Methods
In data analysis and computational mathematics increasing use has been made in re-
cent years of quantities that are characterized by more than two indices. These higher-
order generalizations of vectors (first order) and matrices (second order) are known as
higher-order tensors. Basic tools like the Eigenvalue Decomposition and the Singular
Value Decomposition of matrices can be generalized in several ways to higher-order
tensors. In this talk we will give an overview of the most important tensor decomposi-
tions. We will also explain how these decompositions can be used for factor analysis
and signal separation.  Tensor methods allow one to exploit the intrinsic low dimen-
sionality of signals in different ways.
Philippe Dreesen (KULeuven)
Zero-norm minimization as a polynomial optimization problem
Sparse signal representation and reconstruction are gaining interest in both theoretical
disciplines and application domains such as statistics, machine learning and signal
processing. Such methods provide many improvements over traditional methods. One
prototypical problem can be formulated as finding an n-sparse N-dimensional x such
that Ax = y where A is a prescribed p×N dictionary matrix and y is a p-dimensional
vector of measurements. This is an NP-hard combinatorial problem, hence one solves
the ℓ1 minimization problem in practice, which is computationally more interesting
but may yield sub-optimal results.  In this contribution, we will approach the zero-
14
norm minimization problem from the perspective of polynomial optimization.  The
optimality conditions form a system of polynomial equations.  It can be shown that
the problem of solving a system of polynomial equations can be naturally phrased in
terms of linear algebra, leading to a large eigenvalue problem – moreover, it can be
shown that the global minimizer corresponds to an extremal eigenvalue.  Although
there might not be a direct improvement in terms of computation, this approach sheds
a light on the underlying algebraic-geometric nature of the problem.
Eva Dyer (Rice University)
Sparse Coding in Modular Networks
At every instant, the sensory cortex is bombarded with complex multimodal informa-
tion. To make sense of it all, the brain must process impending sensory stimuli very
efficiently. Investigations into the information processing in the primary visual cortex
(V1) suggest that the visual cortex employs sparsity, both in its population codes (the
number of active neurons is small relative to the size of the population) and spiking
activity. The sparse coding hypothesis was introduced as a means to describe how the
receptive fields of neurons in V1 have adapted over time to the statistics of natural
scenes.  Modern techniques in dictionary learning evolved from this initial investi-
gation into how V1 could learn an optimally sparse representation.  Looking again
to sensory systems for inspiration, we investigate the architecture of the primary vi-
sual area and observed a few remarkable functional properties that appear to emerge
from the connectivity constraints present. We posit that the organization of cells into
densely connected microcircuits or modules, suggests that the visual cortex employs
a coding strategy that produces locally optimal sparse representations instead of re-
quiring denser connectivity to achieve the sparsest global solution.  To find sparse
approximations to images in this constrained setting, we model the system as a union
of subspaces where each image patch can be sparsely represented by a single subspace
in the collection. After motivating our model for sparse coding, we present a method
for learning a union of subspaces that leverages recent results in rank minimization.
Marko Filipovic (Rudjer Boskovic Institute)
Unsupervised segmentation of multispectral images by sparse component analysis
We propose sparse component analysis based approach for unsupervised segmenta-
tion of multispectral images. Segmentation problem is formulated as underdetermined
sparse source separation problem. It is solved in two phases: first, clustering is used to
group image pixels according to their spectral information, wherein cluster centroids
represent spectral profiles of the materials present in the image. Crucial preprocessing
step in this stage is reduction of the number of pixels used in clustering, which dramat-
ically reduces computational cost and in fact enables us to solve the problem. In the
second phase, robust sparse recovery algorithm is used to find representation of every
pixel as a sparse linear combination of spectral profiles found in the first stage. This is
the computationally most demanding part of the algorithm, and could be accelerated
by parallelization, since it is done independently pixel-by-pixel. The effectiveness and
15
robustness of this approach is exemplified on both synthetic and real-world example.
Massimo Fornasier (Austrian Academy of Sciences)
The mathematics of swarming and classical dimensionality reduction principles from
kinetic theory
We review the state-of-the-art in the modelling of the aggregation and collective be-
havior of interacting agents of similar size and body type, typically called swarming.
Starting with individual models based on “particle”-like assumptions, we connect to
hydrodynamic/macroscopic descriptions of collective motion via kinetic theory.  We
emphasize the role of the kinetic viewpoint in the modelling, in the derivation of con-
tinuum models and in the understanding of the complex behavior of the system. We
use the mentioned models for further establishing connections between learning of
dynamical systems modelling collective behavior and sparse recovery.
An extended version of this talk is reported in the book chapter:  “Particle, kinetic,
hydrodynamic models of swarming”, J. A. Carrillo, M. Fornasier, G. Toscani, and
F. Vecil, in “Mathematical modeling of collective behavior in socio-economic and
life-sciences”, Birkh¨auser (in preparation, Eds.  Lorenzo Pareschi, Giovanni Naldi,
and Giuseppe Toscani), 2010.   http://www.ricam.oeaw.ac.at/people/
page/fornasier/bookfinal.pdf
Silvia Gandy (Tokyo Institute of Technology)
A study of multiple measurement approaches for low-rank matrix recovery
In the context of sparse recovery, the problem of recovering a set of jointly sparse
vectors was intensively analyzed.  The topic of this presentation is to generalize the
idea of multiple measurements to low-rank matrix recovery — the recovery of a low-
rank matrix from a small number of linear measurements on the matrix.  The most
natural way to translate the joint support in the vector case to the matrix case is to
require the set of matrices to have the same column and row space.  As the single
measurement case is contained in the multiple measurements setting, there is no hope
to obtain any performance gain in any setting that contains the worst case setting, thus
an average case analysis needs to be considered.
Nicolas Gillis (Universit´e catholique de Louvain)
Dimensionality reduction, classification, and spectral mixture analysis using nonneg-
ative underapproximation
Nonnegative Matrix Factorization (NMF) has recently been successfully used as a
dimensionality reduction technique for identification of the materials present in hy-
perspectral images. In this talk, we present a new variant of NMF called Nonnegative
Matrix Underapproximation (NMU): it is based on the introduction of underapproxi-
mation constraints which enables one to extract features in a recursive way, like PCA,
but preserving nonnegativity. Moreover, we explain why these additional constraints
make NMU particularly well-suited to achieve a parts-based and sparse representation
16
of the data, enabling it to recover the constitutive elements in hyperspectral data. We
experimentally show the efficiency of this new strategy on hyperspectral images asso-
ciated with space object material identification, and on HYDICE and related remote
sensing images.
In collaboration with F. Glineur and R.J. Plemmons.
Bart Goossens (Ghent University)
Solving Various Problems In Image Restoration Through Shearlet-based Regulariza-
tion
Image restoration deals with the recovery of original images from observed degraded
images. Very often, the original images are non-recoverable or very difficult to restore.
Many restoration techniques therefore apply a form of regularization in order to obtain
an acceptable solution. The regularization is usually applied in a transform domain, in
which the images have a sparse representation. In this work, we use the Shearlet trans-
form, which provides a multiresolution analysis (such as the wavelet transform) and is
at the same time an optimally sparse image-independent representation for images that
are smooth away from discontinuities along curves. We consider L1-regularization in
the Shearlet domain and we adopt the Bregman optimization framework to solve the
inversion problem.  We will show that the Bregman framework allows for a lot of
flexibility in practical applications, as the framework inherently decouples the image
model from the degradation model.  We illustrate this through a few very challeng-
ing restoration applications, such as deblurring (deconvolution), demosaicing, signal-
dependent noise removal and bias removal.  We show that our specific design of the
discrete shearlet transform thereby offers a vast performance improvement over total
variation and the discrete wavelet transform, both in visual quality and in PSNR.
Jan Haskovec (Austrian Academy of Sciences)
Mathematical methods for spectral image reconstruction
In old frescos, the visible colour information might be completely or partially lost in
some parts of the painting due to mechanical or (photo)chemical degradation or other
changes of the paint layer.  However, if these reactions do not largely influence the
absorption of the pigments in invisible parts of the spectra (UV and IR), there is a
hope that the original colour information can be faithfully recovered. We demonstrate
how mathematical methods for sparse matrix recovery can be used for this task.  As
shown in [1], the missing data can be exactly reconstructed with very high probability
(i.e., for “almost all” matrices), given only a mild lower bound on the number of sam-
pled entries. Quite recently, two numerical algorithms have been proposed for sparse
matrix recovery:  The singular value thresholding (SVT) algorithm by Cai, Cand`es
and Shen [2], and the iteratively re-weighted least squares minimization (IRWLSM)
by Daubechies, DeVore, Fornasier and G¨unt¨urk [3].  In addition to these two algo-
rithms, which are iterative in nature, we propose a third method (block completion,
BC) for recovery of the missing elements of a low-rank block-shaped matrix, which,
although based on a trivial algebraic manipulation, delivers very competitive results.
17
We demonstrate the performance of these three methods on a sample painting, where
we pick randomly a certain portion (50%) and delete the visible parts of the measured
spectra, while keeping the UV and IR regions. The visible part is then recovered by
application of the three methods mentioned above. We show that the SVT algorithm
typically reaches a relative error of approx.  30% before the convergence drastically
slows down, while the IRWLSM and BC methods usually go down to 10% or even
better.
This is a joint work with W. Baatz (Academy of Fine Arts, Vienna) and M. Fornasier
(RICAM, Linz).
1.  E. J. Cand`es, B. Recht: Exact matrix completion via convex optimization. Foun-
dations of Computational Mathematics, 2009. arXiv:0805.4471v1.
2.  J.-F. Cai, E. J. Cand`es, Z. Shen:  A singular value thresholding algorithm for
matrix completion. October 2008. arXiv:0810.3286.
3.  I. Daubechies, R. DeVore, M. Fornassier, C. S. G¨unt¨urk: Iteratively re-weighted
least squares minimization for sparse recovery.  Commun.  Pure Appl.  Math.,
2009, pp. 35.
Daan Huybrechs (KULeuven)
Fourier series on triangles and higher-dimensional simplices
We describe an original approach to represent smooth functions on a triangular do-
main as classical 2D Fourier series.  The resulting series are amenable to FFT-based
operations for quick evaluation, differentiation and so on.  Though the functions in-
volved are non-periodic (what is a  periodic function on a triangle anyway?),  the
Gibbs-phenomenon is entirely avoided.  Our analysis of the underlying approxima-
tion scheme reveals a natural extension to simplical domains in all dimensions and is
based on the representation theory of finite non-abelian groups.
Marian-Daniel Iordache (University of Extremadura)
Spectral Libraries for Sparse Hyperspectral Unmixing
The hyperspectral unmixing is a problem arising from the relative low spatial res-
olution of the hyperspectral sensors and from the intimate mixture of the materials
present in one pixel at the microscopic level. The hyperspectral data contains mostly
mixed pixels [1], in which there are present at least two materials.  The goal of hy-
perspectral unmixing is to estimate what are the materials present in one pixel (usu-
ally called endmembers), their spectral signatures and their corresponding fractional
abundances (the proportions in which they are present inside the respective pixel).
In a semi-supervised approach, it is supposed that the endmembers belong to a po-
tentially very large collection of spectra, called spectral library.  The advantage of
this method is that the results do not depend on the availability of pure pixels (con-
taining only one endmember) in the scene.  As the number of endmembers present
18
in a pixel is much smaller than the number of spectra contained in the spectral li-
brary, the vector of fractional abundances contains only a few non-zero values, be-
ing sparse.   This characteristic of the solution is explored by the sparse unmixing
techniques, which enforce the sparseness explicitly, by opposition to the non-sparse
ones.  In this paper, we study the influence of the characteristics of the libraries on
the accuracy of the unmixing results, assuming a linear model of the mixtures [2].
Specifically,  we analyse three spectral libraries:  a hypothetical one,  generated as
a collection of spectra containing i.i.d.   Gaussian entries, and two libraries assem-
bled using real signatures from the U.S. Geological Survey (USGS, Available online:
http://speclab.cr.usgs.gov/spectral-lib.html) and the NASA Jet
Propulstion Laboratory’s Advanced Spaceborne Thermal Emission and Reflection Ra-
diometer (ASTER, Available online: http://speclib.jpl.nasa.gov) spec-
tral libraries.  We inspect the internal characteristics of the libraries (namely the mu-
tual coherence and the spectral angle between the signatures), and we compare the
unmixing results obtained using these libraries in a simulated environment, both with
and without enforcing the positivity contraint imposed to the fractional abundances.
We use for our tests a classical non-sparse algorithm: Orthogonal Matching Pursuit
(OMP) [3] and a novel (sparse) one: Sparse Unmixing algorithm via variable Split-
ting and Augmented Lagrangian (SUnSAL) (which exploits the alternating direction
method of multipliers (ADMM) 4 in a way similar to the recent works [5] and [6]),
both in noiseless and noisy environment, showing the high potential that the spectral
libraries have in the sparse unmixing problem.
Joint work with Jos´e M. Bioucas Dias, and Antonio Plaza.
1.  N. Keshava and J. F. Mustard, “Spectral unmixing,” IEEE Signal Processing
Magazine, vol. 19, no. 1, pp. 44–57, 2002.
2.  J. B. Adams, M. O. Smith, and P. E. Johnson, “Spectral mixture modeling: a
new analysis of rock and soil types at the Viking Lander 1 site,” Journal of
Geophysical Research, vol. 91, pp. 8098–8112, 1986.
3.  Y. C. Pati, R. Rezahfar, and P. Krishnaprasad, “Orthogonal matching pursuit:
Recursive function approximation with applications to wavelet decomposition,”
Proceedings of the 27th Annual Asilomar Conference on Signals, Systems and
Computers, Los Alamitos, CA, USA, 2003.
4.  J. Eckstein and D. Bertsekas, “On the douglasrachford splitting method and
the proximal point algorithm for maximal monotone operations,” Mathematical
Programming, vol. 5, pp. 293–318, 1992.
5.  M. Afonso, J. Bioucas-Dias, and M. Figueredo, “Fast image recovery using
variable splitting and constrained optimization,” submitted to the IEEE Trans-
actions on Image Processing, 2009, 2009.
19
6.  ——, “A fast algorithm for the constrained formulation of compressive image
reconstruction and other linear inverse problems,” IEEE International Confer-
ence on Acoustics, Speech, and Signal Processing (ICASSP), Dallas, USA, 2010
(accepted).
Laurent Jacques (University of Louvain)
Dequantizing Non-Uniformly Quantized Compressed Sensing: Weight and See!
Compressed Sensing (CS) aims at reconstructing sparse signals from fewer linear
measurements than the common number of values required by Shannon-Nyquist sam-
pling.  Quantization is however a common communication process that Compressed
Sensing theory has to handle conveniently.  Indeed, any coder or device integrating
a CS encoding, i.e.  the random projections of the signal, has to transmit data to the
decoder in a digital way so as for instance to limit transmission errors.  In this talk,
we present some recent results in this topic when the random measurements of a sig-
nal have undergone a scalar, possibly non-uniform, quantization.  In particular, an
adaptation of a recent technique, called Basis Pursuit DeQuantizer (BPDQ), to non-
uniform scalar quantization of measurements is developed. For uniform quantization,
this algorithm is a convex optimization that minimizes the sparsity of the signal to
reconstruct under a fidelity constraint that is not expressed in the ℓ2-norm anymore
(as for the common Basis Pursuit DeNoising reconstruction) but in the ℓp-norm (with
p ≥ 2). For the non-uniform case, we show that a simple ”weighting” of the ℓp-norm,
with weight related to the quantization bin widths, is sufficient to provide provable
guarantees on the quality of the reconstructed signal.  In short, this quality improves
with p in oversampled situation, i.e. when the number of measurements is higher than
a minimal value increasing with p. The reason of this stability relies on a generaliza-
tion of the Restricted Isometry Property (RIP) to any embedding between two normed
spaces. This property is again respected for with a controlled probability by Random
Gaussian matrices when these two spaces are the weighted-ℓp Banach space and the
ℓ2 Hilbert space, i.e. the case used to prove BPDQ stability. We conclude this talk by
providing some simulations of signal reconstructions on random measurements quan-
tized by the Lloyd’s method in comparison with other state-of-the-art approaches.
Maarten Jansen (K.U.Leuven)
Minimum loss penalties for iterative soft- and hard-thresholding
We present two methods for the estimation of the loss of a threshold operation, namely
Mallow’s Cp and GCV. We observe that minimizing these unbiased estimators intro-
duces a minimization bias, which can be explained as a “mirror effect”.  The mirror
effect can be corrected by replacing hard thresholding by soft thresholding and from
there a corrected Cp or GCV criterion can be established. We then apply GCV to iter-
ative thresholding methods for estimating sparsity in ill conditioned inverse problems.
20
Don Johnson (Rice University)
Signal Processing and Analyzing Works of Art
In examining paintings, art historians use a wide variety of physio-chemical methods
to determine, for example, the paints and ground the artist used.  However, the art
world has been little touched by signal processing algorithms. Our work develops al-
gorithms to examine x-ray images of paintings, not to analyze the artist’s brushstrokes
but to characterize the weave of the canvas that supports the painting.  The physics
of radiography indicates that linear processing of the x-rays is most appropriate. Our
spectral analysis algorithms have an accuracy superior to human spot-measurements
and have the advantage that, through ”short-space” Fourier analysis, they can be read-
ily applied to entire x-rays. We have found that variations in the manufacturing process
create a unique pattern of horizontal and vertical thread density variations in the bolts
of canvas produced. In addition, we can measure the thread angles, providing a way to
determine the presence of cusping and to infer the location of the tacks used to stretch
the canvas in the priming process. We have developed weave matching software that
employs a new correlation measure to find paintings that share canvas weave charac-
teristics.  Using a corpus of over 200 paintings attributed to Vincent van Gogh, we
have found several weave match cliques that we believe will refine the art historical
record and provide more insight into the artist’s creative processes.
Joint work with: C. Richard Johnson Jr. (Cornell University) and Ella Hendriks (van
Gogh Museum).
Felix Krahmer (Universit¨at Bonn)
Lower Bounds for the Error Decay in One-Bit Quantization
Several analog-to-digital conversion methods for bandlimited signals used in appli-
cations, such as ΣΔ quantization schemes, employ coarse quantization coupled with
oversampling. The standard mathematical model for the error analysis of such meth-
ods measures the performance of a given scheme by the rate at which the associated
reconstruction error decays as a function of the oversampling ratio λ. It was recently
shown that exponential accuracy of the form O(2 −r ) can be achieved by appropriate
one-bit Sigma-Delta modulation schemes.  However, the best known achievable rate
constants r differ significantly from the general information theoretic lower bounds.
In particular, no lower bounds specific to one-bit quantization schemes were known.
In this talk we show the first such lower bound for the error decay rate. Our method is
based on large deviation estimates.
This is joint work with Rachel Ward.
Gitta Kutyniok (Universit¨at Osnabr¨uck)
Beyond Wavelets: Compactly Supported Shearlets
Many important problem classes are governed by anisotropic features such as singu-
larities concentrated on lower dimensional embedded manifolds. While the ability to
reliably capture and sparsely represent anisotropic structures is obviously the more
21
important the higher the number of spatial variables is, the principal difficulties arise
already in two spatial dimensions and even there are yet far from being understood.
Three years ago, shearlets were introduced as a means to sparsely encode anisotropic
singularities of 2D data in an optimal way, while – in contrast to previously introduced
directional representation systems – providing a unified treatment of the continuous
and digital world. One main idea is to parameterize directions by slope through shear
matrices rather than angle, which greatly supports the treating of the digital setting.
In this talk, we will first give a general introduction to the theory of shearlets.
Then some very recent results on the construction of shearlet frames generated by
compactly supported shearlets will be highlighted, in particular, showing that these
shearlet frames provide optimally sparse approximations of cartoon-like images.  Fi-
nally, we will discuss applications of shearlet decompositions such as denoising, edge
detection, and data separation.
Gitta Kutyniok (Universit¨at Osnabr¨uck)
Beyond Sparsity: Clustered Sparsity and Data Separation
During the last years, sparsity has become a key concept in applied mathematics.
Methodologies based on sparsity use the fundamental observation that many types of
functions or signals can be very well represented by few non-vanishing coefficients
when they are expanded in a suitable basis or, more generally, in a frame. If a function
or signal does possess such a sparse representation, then it can be recovered from a
small number of linear non-adaptive measurements using ℓ1 minimization.
One application of sparsity methods is the geometric separation of data which is
composed of two (or more) morphologically distinct components. In contrast to other
work on sparsity-driven decompositions, where sparsity of the coefficients plays a role
primarily through the number of nonzeros, this problem complex requires the intro-
duction of the notion of clustered sparsity taking into account the particular clustering
of the significant coefficients.
In this talk, we will first give an introduction to the concept of sparsity-driven de-
compositions and its relation to data separation problems. Then we will focus on some
very recent results in this area, in particular, on the analysis of the separation of point-
and curvelike structures using a wavelet-shearlet dictionary, and on the separation of
components that are clustered sparse within a fusion frame, an extension of frames
which better captures the richness of natural and man-made signals.  Finally, some
applications such as the separation of spines and dendrites in neurobiological imaging
will be discussed.
Wen-shin Lee (University of Antwerp)
Reconstructing a Sparse Trigonometric Polynomial
There are many applications in signal processing and communication systems where
the signals are sparse in some domain such as time, frequency, or space, or alterna-
tively their transform in another domain is sparse. We introduce a new deterministic
(not probabilistic) floating-point algorithm that reconstructs a sparse trigonometric
22
polynomial
p(x1, . . . ,xn ) = X
(j 1 ,…,j n )∈J 1
aj 1 …j n cos(j1×1 + . . . + jnxn )+
X
(j 1 ,…,j n )∈J 2
bj 1 …j n sin(j1×1 + . . . + jnxn ),  J1, J2 ⊂ Z
n
≥0
from a small number of samples.
In exact arithmetic a probabilistic strategy, called early termination, detects the
number of nonzero terms in a multivariate polynomial. In floating-point a number of
heuristics are available. The first sparse interpolation algorithm was given in 1979 by
Zippel. Ben-Or and Tiwari presented their popular algorithm in 1988. Subsequently
Giesbrecht, Labahn and Lee established a connection with Prony’s method. A refor-
mulation of the latter as a generalized eigenvalue problem by Golub, Milanfar and
Varah in 1999 offered a stable numerical algorithm. All of these need to know upper-
bounds for the number of terms in the expression and the “degree” in each variable xi,
in other words, upperbounds for the cardinalities of J1, J2 and bounding boxes in Z
n
≥0
for the index sets J1, J2. A recent discovery [1], based on Rutishauser’s qd-algorithm,
does not suffer from these drawbacks. We explain how it is used for the reconstruction
of a sparse (multidimensional) signal of the form p(x1, . . . , xn ).
Joint work with Annie Cuyt.
1.  A. Cuyt and W.-s. Lee. A new algorithm for sparse interpolation of multivariate
polynomials. Theoretical Computer Science, 409(2):180–185, 2008.
David Mary (University of Nice)
Properties of thresholding functions for different sparsity-based denoising scenarios
The Bayesian Maximum A Posteriori (MAP) estimation requires minimizing in x a
functional of the form J(x) = Q(x, y) + Φ(x), where Q(x, y) is a data fidelity cri-
terion – related to the noise model; and Φ(x) is a penalization term – related to the
probability density function of the unknown vector x, which is defined a priori.
In this framework, particular prior densities are known to yield sparse estimates
of x.  This property is interesting for denoising, compression, pattern recognition or
inpainting applications, as natural signals tend to have sparse representations in some
transform domains. Examples of such priors are the Laplacian, and more generally the
generalized Gaussian densities, which lead respectively to penalizations of the form
of the ℓ1 norm, and of the ℓp
p
pseudo-norm.
This work proposes a review of results regarding the properties of the MAP esti-
mates which are produced by various penalization functions and fidelity criteria. Prop-
erties of interest are in particular the existence of a thresholding effect, the continuity
of the MAP estimate, or equations verified by the threshold point.
23
Jonas Offtermatt (University of Stuttgart)
An adaptive algorithm for providing sparse solutions in an application of systems bi-
ology
We approximate sparse solutions of inverse problems by an adaptive discretization
scheme, where the adaptivity is controlled by cheaply computable refinement and
coarsening indicators. These indicators are based on works by Ben Ameur, Chavent,
Jaffre and Kaltenbacher for estimating a distributed hydraulic transmissivity. In Sys-
tems Biology a recent sparse problem is the reconstruction of gene networks out of
micro array data sets. This is in general a highly underdetermined, ill-posed and sparse
problem, but can be modeled in a simplified form as a linear inverse problem. We ap-
plied our discretization scheme to this linear model and compared the results, with the
well known iterative soft thresholding algorithm.
Ljiljana Platisa (UGent)
Image Blur Estimation Based on the Average Cone of Ratio in the Wavelet Domain
Recently, we proposed a new algorithm for objective blur estimation using wavelet
decomposition. In particular, the method makes use of a wavelet domain local regu-
larity measure named average cone ratio (ACR). The ACR quantifies joint expansion
of the magnitudes of wavelet coefficients inside a cone of influence which is centered
at the given spatial position and it was shown to be a good estimate of the local Lip-
schitz exponent. Moreover, it has been shown that the ACR measure is highly robust
to noise.
In our work, these advantageous properties of ACR: estimating local edge reg-
ularity while being insensitive to noise, are used for blur estimation.  Namely, our
proposed method is designed to estimate blur as a function of the center of gravity of
the ACR histogram. The new metric is named CogACR. The method of CogACR is
applicable both in case where the reference image is available and when there is no
reference.
Our results demonstrate a consistent performance of the proposed metric for a
wide class of natural images and in a wide range of both Gaussian and out of focus
blurriness, and over a wide range of noise.  One of the important conclusions of our
study is that the CogACR may be used as a powerful metric of image blur over a wide
range of blur levels and while being nearly insensitive to noise. Moreover, the results
of the study prove the ability of the new blur metric to clearly discriminate between
low level quanta of blur in both reference- and no-reference case which encourages
further research of the subject. Compared to the existing state-of-the-art methods for
blur estimation, the CogACR metric demonstrates not only the significantly greater
robustness to noise but also the ability to discriminate between different levels of blur
in the image for a noticeably wider range of blurriness.  Most recently, we have in-
vestigated the feasibility of using CogACR as a real-time metric in HD video qual-
ity assessment.  Encouragingly, those results suggest that, despite its computational
complexity, the metric can be efficiently implemented on a commercially available
processor and achieve real-time performance for HD inputs.
24
Nevertheless, some important aspects of CogACR metric remain to be addressed
in order for the method to become a practically working one. Some of these aspects
include content sensitivity and this is exactly the focus of our current research interest
and a topic of our ongoing investigations.
Javier Portilla (Instituto de Optica, CSIC)
Sparse Approximation: A general discussion and a simple algorithm
•   What do we mean by ”sparse” representations?
•   Differences between the analysis (linear) and the generative (non-linear) sense.
•   Sparse approximation: problem formulation using a (quasi)norm ℓp.
•   Advantages and disadvantages of using a convex norm (p ≥ 1).
•   ℓ0-AP: A simple optimization method based on alternated projections, with Parse-
val frames.
Javier Portilla (Instituto de Optica, CSIC)
Efficient ℓ0-based sparse approximation using Parseval frames
•   Two alternative, quasi-equivalent, ways to do local ℓ0-optimization, both resulting
in Iterated Hard Thesholding.
•   Gradual de-smoothing of the cost function gives raise to Dynamic Iterated Hard
Thresholding.
•   Doubly quasi-asymptotically optimal behavior of DIHT.
•   A variant of the method exactly fulfilling the NAP (Nested Approximation Prop-
erty).
•   Comparison to other sparse approximation methods.
Javier Portilla (Instituto de Optica, CSIC)
From approximation to estimation. Some image processing examples.
•   Promoting synthesis-sense vs. analysis-sense sparsity.
•   A posteriori deterministic degradations: quantization, lost pixels.
•   Formulation using a consistency set.
•   Non-localized degradations: noise and blur plus noise.
•   The ℓ0-AbS method.
Holger Rauhut (University of Bonn)
Sparse Legendre expansions via ℓ1-minimization
We consider the recovery of polynomials that are sparse with respect to the basis of
Legendre polynomials from a small number of random sampling points.  We show
that a Legendre s-sparse polynomial of maximal degree N can be recovered from
m  ≍  s log
4
(N) random samples that are chosen independently according to the
Chebyshev probability measure π −1 (1 − x2 ) −1/2dx on [−1, 1].  As an efficient re-
covery method ℓ1-minimization can be used.  Our results extend to a large class of
orthogonal polynomial systems on [−1, 1]. As a byproduct, we obtain condition num-
25
ber estimates for preconditioned random Legendre matrices that should be of interest
on their own.
This is joint work with Rachel Ward.
Joseph Richards (Carnegie Mellon University)
Sparse Prototyping for Astrophysical Spectra
Astronomical spectral databases such as the Sloan Digital Sky Survey (SDSS) are
populated with noisy, high-dimensional emission spectra from millions of galaxies.
For each galaxy, there is a complicated relationship between its physical properties and
its observable spectrum, which consists of information across thousands of wavelength
bins. This inverse problem can only be solved using physically-motivated simulation
models.  As is typical, however, the information in each spectrum that is useful for
estimating galaxy properties resides in a low-dimensional subset of the full, high-
dimensional space. In Richards et al. (2009), we performed parameter estimation for
galaxies using a novel approach to sparse prototyping based on exploiting the low-
dimensional diffusion map parametrization of spectra.  Specifically, the goal of our
study is to estimate the star formation history (SFH) of each galaxy in a database by
fitting its spectrum as a mixture of a small number of simple stellar population (SSP)
prototypes.   We show that a carefully-chosen basis of SSP prototypes selected by
diffusion k-means yields better SFH estimates than the SSP bases used in the literature
and bases derived from other methods. I will discuss the results from this study and
present diffusion k-means as a method of sparse prototyping in a general setting.
Andrea Rueda-Olarte/Eduardo Romero         (Universidad Nacional de Colombia)
Super-Resolution of Brain MR Images based on Sparse Representations
The problem of Partial Volume (PV) in Magnetic Resonance (MR) images is due to
the limited spatial resolution of the MR equipment, causing blurring at the interfaces
between tissues.  Together with noise and intensity inhomogeneities, the PV effect
affects the precision of the tissue segmentation task, producing misclassification of
voxels and distorting quantitative results of posterior morphometry tasks such as cor-
tical thickness estimation. This paper presents the application of an already-proposed
methodology for super-resolution of images, based on sparse representations, that at-
tempts to overcome the PV effect by generating high-resolution versions of brain MR
images.
Khalid Sabri (LATT-CNRS)
Efficient implementation of greedy algorithms for sparse images deconvolution
We are interested in the use of Sparse representations for images deconvolution. In this
study, we focus on the implementation of some greedy algorithms like OMP (Orthog-
onal Matching Pursuit), OLS (Orthogonal Least Square), StOMP (Stage-wise OMP)
and ℓ2 − ℓ0 minimisation by the SBR (Single Best Replacement) in the image de-
convolution framework. The key idea is to replace matrix computation operations by
26
convolution operations. An efficient implementation of such idea, accounting for the
sparsity of the images, could dramatically reduce the required memory storage and
computational burden. Let H be the linear operator associated to the convolution be-
tween the input image x of size Lx × Lx  and the point spread function (PSF) h of
size Lh × Lh. The corresponding Toeplitz-Bloc-Toeplitz matrix H is of size L2
x × L2
x
(actually, non square matrix can be considered depending on how the edge conditions
are handled). As the PSF is generally of small dimension compared to the images, H
is only filled with L2
x × L2
h
non-zero elements. Nevertheless, with convolution oper-
ation implementation, there is no need to neither built H nor to store it, only the PSF
has to be stored; each pixel (n, m) of the resulting image y is iteratively given by the
products of elements of h with the pixels of x in the neighborhood of (n, m). In this
work, we present an efficient implementation of each steps of the greedy algorithms
OMP, OLS, StOMP and SBR, using such an approach, in the case of large data size.
The implementation is studied for the various edge conditions (equivalent to the shape
options full, same and valid of the Matlab conv2function).
Joint work with Herv´e Carfantan.
Aswin Sankaranarayanan (Rice University)
Compressive acquisition of dynamic scenes
We address the problem of compressive acquisition of dynamic scenes. The ephemeral
nature of time and temporal sequences makes this a challenging problem. Much of ex-
isting methods for dynamic scenes work under the assumption of a snapshot imager,
that provides multiple time-synchronous compressive measurements at every instant.
Typically, the number of measurements taken at a particular time instant are, by them-
selves, sufficient to reconstruct the image.  The temporal smoothness of the visual
signal only helps in improving the robustness. Unfortunately, such an approach does
not extend to the paradigm of the single pixel camera (SPC), which has both practical
and philosophical elegance.  In this talk, we propose algorithms for recovering dy-
namic scenes from a SPC. Our method relies on the idea of motion compensation, to
identify lower dimensional structures underlying the temporal signal. In particular, we
model the scene as a linear dynamical system, and use low rank properties inherent to
such systems. The use of rich motion models such the linear dynamical system allows
us to use measurements taken in ”future” to constrain those in ”past”.  We highlight
the use of such methods in video recovery, high speed imaging, and dynamic textures.
This is joint work with Richard Baraniuk at Rice University.
Marco Signoretto (K.U. Leuven)
High Dimensional Sparse Estimation with Multiple Graphs
High dimensional linear modeling deals with the problem of estimating coefficient
vector β ∈ R
p
based on a sample of size n when p is large and, possibly, n ≪ p. A
general approach prescribes to use penalized empirical risk minimization to find an
estimate
ˆβ.  From a theoretical standpoint, consistency can be usually proved based
on certain structural assumptions on the generative model β⋆
.  In particular, the er-
27
ror  ˆβ − β⋆
can be related to how well the underlying structure is captured by the
penalty function r(β) (see e.g.  [1] and references therein).  In the high dimensional
setting, a popular example of structural assumption is sparsity and the corresponding
choice of the penalty is r(β) = kβk 1. More recently, structure-inducing norms have
been proposed as a promising alternative [9,11,12].  The general idea is to convey
structural assumption on the problem, such as grouping or hierarchies over the set of
input variables, by suitably crafting the penalty.  A different but related method re-
quires to endow the set of covariates with a graph structure [6].  In this case nodes
are variables, edges represent interactions and groups naturally emerge as connected
components of the graph. Given the graph Laplacian L one then uses the composite
penalty rα (β) = (1 − α)hβ, Lβi+ αkβk 1 for some user-defined 0 < α < 1. The idea
extends the elastic-net [8] to endow prior information synthesized in the graph. The
first term in the penalty enforces smooth profile of coefficients associated to neighbor-
ing nodes.  The second term ensures that coefficients with small contribution would
shrink to exact zero. In this contribution we study the penalized estimator arising from
the use of rα as the penalty function and discuss it in connection with the other afore-
mentioned penalties.  We show theoretical properties of the corresponding penalized
estimator under specific structural assumption on β⋆
.  We then illustrate a recently
proposed extension [4] which allows to combine multiple graphs estimated from data.
We finally demonstrate the effectiveness of the procedures in a number of case studies.
Joint work by Marco Signoretto and Johan A.K. Suykens.
1.  S. Negahban, P. Ravikumar, M.J. Wainwright and B. Yu. A unified framework
for high-dimensional analysis of M-estimators with decomposable regularizers.
Advances in Neural Information Processing Systems, 2009.
2.  F.R.K. Chung. Spectral graph theory. American Mathematical Society, 1997.
3.  Y. Kim, J. Kim and Y. Kim.  Blockwise sparse regression.  Statistica Sinica,
16(2):375, 2006.
4.  M. Signoretto, A. Daemen, C. Savorgnan, J.A.K. Suykens. Variable Selection
and Grouping with Multiple Graph Priors. NIPS Workshop on Optimization for
Machine Learning, 2009.
5.  E. D. Kolaczyk.  Statistical Analysis of Network Data:  Methods and Models.
Springer, 2009.
6.  C. Li and H. Li. Network-constrained regularization and variable selection for
analysis of genomic data. Bioinformatics, 24(9):1175–1182, 2008.
7.  M.R. Osborne, B. Presnell and B.A. Turlach.   On the LASSO and Its Dual.
Journal of Computational and Graphical Statistics, 9(2):319–337, 2000.
8.  H. Zou and T. Hastie. Regularization and variable selection via the elastic net.
J. Roy. Stat. Soc. Ser. B, 67(2):301–320, 2005.
28
9.  P. Zhao, G. Rocha, B. Yu. The composite absolute penalties family for grouped
and hierarchical variable selection. Annals of Statistics, 37:3468–3497, 2009.
10.  P. Bickel,  Y. Ritov and A. Tsybakov.   Simultaneous analysis of Lasso  and
Dantzig selector. Annals of Statistics, 37(4):1705–1732, 2009.
11.  M. Yuan and Y. Lin. Model selection and estimation in regression with grouped
variables. J. Roy. Stat. Soc. Ser. B, 68(1):49–67, 2006.
12.  R. Jenatton, J.Y. Audibert, F. Bach. Structured variable selection with sparsity-
inducing norms. Technical report, arXiv:0904.3523, 2009.
David Stork (Ricoh Innovations)
Did the great masters “cheat” using optics? Computer science, optics and art history
confront a bold theory
In 2000, the artist, photographer and set designer David Hockney made the bold and
highly promoted claim that some early Renaissance painters, as early as 1430, secretly
built optical projectors, projected images onto their supports (canvas, oak panel, paper,
…) and traced such images, and that this procedure was a key source in the rise in real-
ism in the ars nova or “new art” of that time. The earliest work adduced as evidence for
this claim is Robert Campin’s M´erode altarpiece, executed in Bruges Belgium. This
talk will review the claim and the scholarly publications and technical image analyses
of works by Jan van Eyck, Robert Campin, Lorenzo Lotto, Hans Memling and oth-
ers, from eleven scientists, nine historians of optics and art, addressing this projection
claim. The conclusion of these international, independent scholars is unanimous. You
will never see early Renaissance works the same way again.
David Stork (Ricoh Innovations)
Computer graphics in the history and interpretation of art: Computer science, optics
and art history confront a bold theory
Computer graphics reconstructions of realist artists’ studios and tableaus allow schol-
ars to explore ”what if” scenarios and better understand the working methods of such
artists.  Problems such as the location and number of illuminants, the geometry and
scale of objects, the nature of the sightlines, the perspective coherence and the focal
length of putative projectors are all add This talk will illustrate the power of these new
techniques through analyses of works by Jan van Eyck, Hans Memling, Georges de la
Tour, Parmigianino, Caravaggio and Diego Vel`azquez.
Maja Temerinac-Ott (University of Freiburg)
Multichannel Image Restoration Based on Optimization of the Structural Similarity
Index
In this paper a framework for multichannel image restoration based on optimization of
the structural similarity (SSIM) index is presented. The SSIM index describes the sim-
ilarity of images more appropriately for the human visual system than the mean square
29
error (MSE). It has not yet been explored for the multi channel restoration task. The
construction of an optimization algorithm is difficult due to the non-linearity of the
SSIM measure. The existing solution based on a quasi-convex problem formulation is
successfully extended for the multichannel image restoration. The correctness of the
algorithm is verified on sample images and it is shown that multi-view information
can significantly improve the restoration results.
Stefanie Tenorth (University of Duisburg-Essen)
A Hybrid Algorithm for Image Approximation Based on the EPWT
The EPWT is a locally adaptive wavelet transform that makes use of strong correla-
tions of adjacent pixels. First, a path through all pixels (i.e. a permutation of all pixels)
is calculated, so that pixels with similar values are adjacent on the path. Then a one-
dimensional wavelet transform (e.g.  the biorthogonal Cohen-Daubechies-Feauveau
9/7-transform) is applied to the newly ordered pixels.  However, the EPWT suffers
from its adaptivity costs that arise from the storage of path vectors.
The proposed hybrid algorithm exploits the advantages of the usual tensor prod-
uct wavelet transform for the representation of smooth images and uses the Easy Path
Wavelet Transform (EPWT) for an efficient representation of edges and texture.  It
works as follows. The given image is smoothed by a diffusion procedure. A biorthog-
onal tensor product wavelet transform is applied to the smoothed image. Further, the
EPWT is used to construct a sparse representation of the (shrunken) difference image.
Numerical results show the efficiency of this procedure.
This talk is based on joint work with Gerlind Plonka (University of Duisburg-Essen,
Germany) and Daniela Ros¸ca (Technical University of Cluj-Napoca, Romania).
George Tzagkarakis (University of Crete & Foundation for Research and
Technology-Hellas (FO.R.T.H.))
Bayesian Compressed Sensing using Alpha-Stable Distributions
The majority of previous compressed sensing (CS) techniques for the reconstruction of
a sparse signal solve constrained optimization problems. Commonly used approaches
are typically based on convex relaxation (Basis Pursuit (BP), Least Absolute Shrink-
age and Selection Operator (LASSO)), non-convex (gradient based) local optimization
(Reweighted ℓ1 minimization) or greedy strategies ((Orthogonal) Matching Pursuit
((O)MP). All these methods have been applied in scenarios where the underlying pro-
cess generating the signal and/or the noise follows a Gaussian model.  However, the
normality assumption is violated in several distinct environments, such as in under-
water acoustics, in sonar/radar and in finance, where the associate signals take large-
amplitude values much more frequently than what a Gaussian model implies. Several
studies have proposed that the statistics of many highly impulsive, and thus highly
sparse, signals can be modeled with high accuracy by means of the family of alpha-
Stable distributions and specifically of the sub-class of symmetric alpha-Stable (SaS)
distributions. Despite the power of the SaS model, the lack of closed-form expressions
for all but a few stable distributions (Gaussian, Cauchy) has been a major drawback to
30
their use by the signal processing community. Thus, the main contribution of our work
is that it establishes the first connection between the two fields, namely, the theory of
CS and the SaS models.  For this purpose, we propose an iterative greedy algorithm
for CS sparse approximation of impulsive signals corrupted by additive heavy-tailed
noise.  By noting that the problem of sparse representation reduces to a problem of
determining a sparse basis configuration we propose a basis selection rule which is
best adapted to the true underlying heavy-tailed behavior of the sparse signal, by em-
ploying a statistical measure of the distance between two SaS vectors defined in terms
of the so-called Fractional Lower-Order Moments (FLOMs).   Moreover, we intro-
duce a matrix (measurement basis) for acquiring the measurements, which is also best
adapted to the true statistical characteristics of the sparse signal. This is in contrast to
the previous CS methods which are non-adaptive, in the sense that the elements of the
measurement basis are drawn from a predetermined probability distribution without
taking into consideration the underlying statistics of the given sparse signal. Finally,
the proposed SaS-CS algorithm for sparse approximation of a highly impulsive signal
is extended to a distributed framework by solving iteratively a constrained optimiza-
tion problem using the duality theory and the method of sub-gradients.
Raf Van de Plas (K.U.Leuven)
Wavelet approaches to enable the exploration of organic tissue via multivariate anal-
ysis of mass spectral imaging data
Abstract: In recent years the molecular mechanisms underlying many of the world’s
most pressing diseases have been shown to exhibit a complexity that exceeds most of
the early expectations. The searches for disease markers from indirect sources such as
blood or plasma have known reasonable success, but are largely aimed at diagnostic
efforts. A more fundamental understanding of the pathomechanisms underlying a par-
ticular disease requires an examination closer at the source by investigating disease-
related tissue directly. As more biopsy material and model organisms become avail-
able for pathologies ranging from cancer to diabetes, there is an increased demand for
analytical methods capable of delivering insight into the molecular content of organic
tissue. One of the most promising technologies in the study of disease-related tissue
is mass spectral imaging.
Mass spectral imaging (MSI) is a developing technology that adds a spatial dimen-
sion to mass spectrometry-based biochemical analysis. It enables researchers to study
the spatial distribution of biomolecules such as proteins, peptides, and metabolites
throughout organic tissue sections.  MSI has particular merit in exploratory settings
where there is no prior hypothesis of relevant target molecules. It is rapidly becoming
a potent exploratory instrument for tissue biomarker studies.
MSI is a high-throughput technique that mines massive amounts of measurements
from a single tissue section. It is not uncommon for a single MSI experiment to de-
liver a data set that exceeds several gigabytes of raw measurements. Many MSI studies
currently use only a fraction of the information available in these vast data sets. They
focus on the spatial distribution of only one or two known molecules, while ignoring
31
information from the hundreds or thousands of other molecules that were measured.
In an effort to unlock the vast potential of MSI, there is an increased push towards
data-mining these data sets using various forms of supervised and unsupervised mul-
tivariate techniques (e.g.  classification, clustering,…).  Initial results have been very
promising, delivering new insights into diseases such as diabetes and amyotrophic lat-
eral sclerosis. However the size of the data sets combined with the scalability of the
algorithms often causes these data-mining efforts to be practically inefficient or even
infeasible. As various parameters such as the covered tissue surface area, the spatial
resolution, and the extent of the mass range grow, MSI data sets rapidly become very
large, making analysis from a computational and memory standpoint increasingly dif-
ficult.  In this work we employ the discrete wavelet transform (DWT) as a means of
reducing the dimensionality of the data, while retaining a maximum amount of bio-
chemical information. The DWT is used to deliver a more compact description of each
mass spectrum, expressed as wavelet coefficients.  The goal is to perform the multi-
variate data-mining analyses in the reduced wavelet space rather than directly on the
raw data. This enables us to keep a maximum of chemical information concurrently
available for the algorithms, while at the same time severely reducing the memory and
computational footprint of the operation.
We demonstrate the efficacy of the approach on a sagittal section of mouse brain as
well as on a case study of rat testes. The massive reductions in required memory and
computation time, while obtaining identical results, are illustrated using unsupervised
trend detection via principal component analysis (PCA) and using a typical super-
vised classification scenario. This work finds itself at the crossroads between biology,
physics, and machine learning and it introduces wavelets as a means of unlocking the
potential of a novel molecular imaging technology known as mass spectral imaging.
Joint work with Bart De Moor (Katholieke Universiteit Leuven),  and Etienne
Waelkens (Katholieke Universiteit Leuven).
Pierre Vandergheynst (Ecole Polytechnique F´ed´erale de Lausanne)
Wavelets on graphs via spectral theory
We propose a novel method for constructing wavelet transforms of functions defined
on the vertices of an arbitrary finite weighted graph. Our approach is based on defin-
ing scaling using the the graph analogue of the Fourier domain, namely the spectral
decomposition of the discrete graph Laplacian Ł.  Given a wavelet generating kernel
g and a scale parameter t, we define the scaled wavelet operator T
t
g = g(tŁ).  The
spectral graph wavelets are then formed by localizing this operator by applying it to
an indicator function.  Subject to an admissibility condition on g, this procedure de-
fines an invertible transform. We explore the localization properties of the wavelets in
the limit of fine scales. Additionally, we present a fast Chebyshev polynomial approx-
imation algorithm for computing the transform that avoids the need for diagonalizing
Ł. We highlight potential applications of the transform through examples of wavelets
on graphs corresponding to a variety of different problem domains.
32
Pierre Vandergheynst (Ecole Polytechnique F´ed´erale de Lausanne)
Spread spectrum imaging techniques in MRI and Radio-interferometry: experimental
promises
We explore how ideas neatly linked with compressed sensing can be applied to two
problems where under-sampling is performed in the Fourier domain: MRI and Radio-
interferometry.   In particular,  we show that in  both cases a spread-spectrum pre-
modulation of the signal yields far better reconstruction than classical techniques.
We discuss how this modulation is imposed by the physics of the problem in inter-
ferometry and how it can be practically implemented in MRI. Finally we highlight
the universality of these results under the light of compressed sensing theory, using
simple incoherence arguments.
Dimitri Van De Ville (EPFL and University of Geneva)
Surfing the Brain: Wavelets and Sparsity for Functional Brain Imaging
Functional human brain mapping using magnetic resonance imaging is playing an
increasingly important role in neuroscience, biology and medicine.  Advanced tools
from signal processing and statistics are needed to fully exploit the potential of the
large and complex datasets at hand.  Multiscale and sparsity are two important con-
cepts to nurture further advances in this field.
Traditional functional MRI data analysis typically tests for the presence of a hy-
pothetical task-related blood oxygen-level dependent (BOLD) response.  If such ev-
idence is found (on statistical grounds), voxels are declared as ”active”.  The most
popular statistical framework is based on the general linear model (GLM) with regres-
sors of interest and other variates.  However, the huge number of univariate tests re-
quires proper correction for multiple comparisons. The most widely deployed method
uses spatial Gaussian prefiltering to obtain sufficient smoothness such that continu-
ous Gaussian random field theory and Euler characteristics can be applied to yield
lower statistical threshold lower than the one obtained with conservative Bonferroni
correction.
In the first part of this talk, I will explain how the (spatial) wavelet transform can
circumvent the need of spatial smoothing while still guaranteeing strong control of
false positives without loss of sensitivity compared to smoothing approaches.  The
core of the framework relies on a theorem that bounds the null hypothesis rejection
probability after reconstruction from thresholded wavelet coefficients.
In the second part of this talk, I will show that (temporal) wavelets can be de-
signed specifically for the hemodynamic system that governs the BOLD response.
These wavelets have exponential vanishing moments, which means that they annihi-
late exponential polynomials.  Proper tuning of the wavelet’s underlying differential
operator can then lead to sparse representation of hemodynamic signals.  The use of
ℓ1 minimization then allows to recover activation-related signals, as demonstrated by
synthetic and real-world data.
1.  D. Van De Ville, M.L. Seghier, F. Lazeyras, T. Blu, M. Unser, ”WSPM: Wavelet-
33
Based Statistical Parametric Mapping,” NeuroImage, vol. 37, no. 4, pp. 1205-
1217, October 1, 2007.
2.  D. Van De Ville, T. Blu, M. Unser, ”Integrated Wavelet Processing and Spatial
Statistical Testing of fMRI Data,” NeuroImage, vol. 23, no. 4, pp. 1472-1485,
December 2004.
3.  I. Khalidov, D. Van De Ville,  J. Fadili, F. Lazeyras, M. Unser, ”Activelets:
Wavelets for Sparse Representation of Hemodynamic Responses,” submitted.
4.  WSPM toolbox: http://miplab.epfl.ch/wspm/
Dimitri Van De Ville (EPFL and University of Geneva)
Steerable Wavelet Pyramids and Reconstruction from a Compact Multiscale Primal
Sketch
I will show the tight link between wavelets, some classical image processing opera-
tors and David Marr’s theory of early vision. The starting point is a general wavelet
basis design procedure. Starting from fundamental invariances principles, we identify
the corresponding class of operators and the associated Green’s functions.  Next, we
design a single-generator wavelet, in an analytical way, and we show that it yields a
semi-orthogonal basis of L2 ( R 2), irrespective of the dilation matrix used. Moreover,
the wavelet behaves as the chosen operator when applied to data. We also provide an
efficient FFT-based filterbank implementation. The design procedure is illustrated for
operators that are translation-, scale- and rotation-invariant, which relates to the (iter-
ated) Laplacian. The isotropic polyharmonic B-spline wavelets are ideally suited for
the analysis of multidimensional data with fractal characteristics (isotropic differentia-
tion and whitening property) or for applications such as statistical resampling without
directional bias. Next, we consider the rotation covariant case that leads to the class of
complex gradient-Laplace operators.  Through generalized polyharmonic B-splines,
we are able to construct complex-valued steerable wavelets; i.e., a suitable combina-
tion of real and imaginary part of the wavelet coefficients allows to obtain the response
for any angle. We show how the transform’s properties can be improved by introduc-
ing a pyramid with mild redundancy. The steerable polyharmonic wavelets has some
striking similarities with Marr’s theory of vision.  We show an image reconstruction
algorithm based on multiscale edge charactization only.
Sabine Van Huffel (Katholieke Universiteit Leuven)
Tensor-based biosignal processing
This contribution deals with tensor decompositions and their benefits in biomedical
signal processing.   After a brief introduction of tensors,  different decompositions
are described and their computational algorithms are introduced as generalizations
of matrix-based counterparts. In particular, we will focus on ’Parallel Factor Analy-
sis’ (Parafac, also known as Candecomp or the CP model), the most popular tensor
34
decomposition, and overview its mathematical properties and present some variants
by adding constraints.  The CP model decomposes in a unique way a higher-order
tensor in a minimal sum of rank-1 ’atoms’.  Furthermore, we will give an overview
of biomedical applications of these algorithms and their benefits will be illustrated in
a variety of case studies.  In particular, we will focus on the presurgical evaluation
of refractory partial epilepsy for the delineation of the irritative and ictal onset zones
using long-term electroencephalographic (EEG) recordings.
Keywords: tensor, PARAFAC, biomedicine, signal processing, multilinear algebra
Nico Verbeeck (K.U. Leuven)
Improved Wavelet Analysis of Mass Spectral Imaging Data for Feature Selection and
Data Compression through Incorporation of Spatial Information
Mass Spectral Imaging is a relatively new molecular imaging technology that makes
it possible to detect thousands of molecules throughout tissue simultaneously, ranging
from low-mass metabolites to high-mass proteins. This technology is of prime interest
for the molecular characterization of tissue in biomedical studies. In recent years, MSI
data sets have grown in size to such extent that it becomes more and more infeasible to
computationally analyze them in their raw form due to both memory and calculation
time constraints.  Additionally, traditional datamining algorithms are unable to cope
with the high dimensionality of the data.  As a result, there is a growing need for
thorough preprocessing steps, which can reduce both the dimensionality and data size
through proper feature selection.
Research at ESAT by Van de Plas et al.  has shown solid results using Discrete
Wavelet Transform (DWT) on mass spectra to perform feature selection, thus reduc-
ing data size, dimensionality and noise. Our newest method further improves on this
approach by incorporating one of the key aspects of MSI, spatial information, to better
determine what part of the data can truly be considered noise.  By using this infor-
mation, we can selectively retain only those detail components that exhibit a spatial
structure.   This spatial structure is currently recovered using Principal Component
Analysis.
We demonstrate the performance of this new compression method on a case study
of a sagittal section of mouse brain and compare the results to the Van de Plas et al.
method and to direct analysis of the raw measurements.
Joint work with Raf Van de Plas, Bart De Moor and Etienne Waelkens (KULeu-
ven).
Sergey Voronin (Princeton University)
ℓ1-regularization for applications
There are multiple applications of ℓ1-optimization, a sparsity inducing regularization
technique.  In our project arising in Global Seismic Tomography, we are faced with
the problem of reconstructing sparse solutions in the wavelet domain from very large
under-determined linear systems of equations.  We use recently developed iterative
thresholding techniques which utilize simple linear operations and yield well to paral-
35
lel computation. We discuss these issues along with possible approximation strategies
and show some results from computations.
Mariya Zhariy (Johannes Kepler University)
Truncated Soft Shrinkage Iteration with Discrepancy Based Stopping Rule. Applica-
tion to Inverse Problems with Data Noise.
The main goal of the current paper is to define a fast regularization approach for lin-
ear ill-posed inverse problems based on the sparse approximation techniques.  It is
well known that the combination of the least squares minimization with an ℓp-penalty
term assures stable approximation of the solution in the ill-posed case. The minimizer
of the ℓp-optimization problem are usually computed iteratively. As an example, we
consider the soft-shrinkage iteration. For the estimation of the optimal regularization
parameter in case of the noisy data, an additional iteration over the parameter values
is needed. In our case, we use the discrepancy principle in order to adjust the values
of the regularization parameter during the soft-shrinkage iteration.  The sparsity ap-
proach can be refined by introducing sparse operator approximation. The reduction of
the computational effort is performed on two stages. On the one hand, the application
of the soft-shrinkage approach reduces the number of calculated coefficients per iter-
ation. On the other hand, the regularization parameter search is incorporated into the
minimization procedure, which allows to avoid an extra loop over the regularization
parameter values.
36
4    List of participants
Jan Aelterman
Ghent University
jan.aelterman@telin.ugent.be
Reema Al-Aifari
Johannes Kepler University Linz
al-aifari@indmath.uni-linz.ac.at
Musa Alrefaya
Vrije Universiteit Brussel
malrefay@etro.vub.ac.be
Carlos Alzate
Katholieke Universiteit Leuven
carlos.alzate@esat.kuleuven.be
Jean-Pierre ANTOINE
UCL
jean-pierre.antoine@uclouvain.be
Stephan W. Anzengruber
Radon  Institute  for  Computational  and
Applied Mathematics
stephan.anzengruber@oeaw.ac.at
Ulas Ayaz
Bonn University
ulasayaz@yahoo.com
Samir Kumar Bhowmik
University of Amsterdam
S.K.Bhowmik@uva.nl
Ismael Rodrigo Bleyer
Doktoratskolleg ”Computational Mathe-
matics”
ismael.bleyer@dk-compmath.jku.at
Florian Boßmann
University Duisburg-Essen
florian.bossmann@uni-due.de
S´ebastien Bourguignon
Universit´e de Nice
Sebastien.Bourguignon@oca.eu
Philippe Cara
Vrije Universiteit Brussel
pcara@vub.ac.be
Catherine Charles
ULg, Gembloux Agro Bio Tech
C.Charles@ulg.ac.be
Nabiollah Godarzvand Chegini
Korteweg-de  Vries  institute  for  mathe-
matics
nabichegini@yahoo.com
Bruno Cornelis
Vrije Universiteit Brussel
bcorneli@etro.vub.ac.be
Annie Cuyt
University of Antwerp
annie.cuyt@ua.ac.be
Christophe Damerval
Joint  Research  Center  of  the  European
Commission
christophe.damerval@jrc.ec.europa.eu
Ingrid Daubechies
Princeton University
ingrid@math.princeton.edu
37
Michel Defrise
Vrije Universiteit Brussel
mdefrise@vub.ac.be
Lieven De Lathauwer
K.U.Leuven
lieven.delathauwer@kuleuven-kortrijk.be
Veronique Delouille
Royal Observatory of Belgium
v.delouille@oma.be
Christine De Mol
Universit´e Libre de Bruxelles
demol@ulb.ac.be
Philippe Dreesen
KULeuven
philippe.dreesen@esat.kuleuven.be
Ann Dooms
Vrije Universiteit Brussel
adooms@vub.ac.be
Eva Dyer
Rice University
e.dyer@rice.edu
Marko Filipovic
Rudjer Boskovic Institute
filipov@irb.hr
Massimo Fornasier
Austrian Academy of Sciences
massimo.fornasier@oeaw.ac.at
Silvia Gandy
Tokyo Institute of Technology
gandy@comm.ss.titech.ac.jp
Nicolas Gillis
Universit´e catholique de Louvain
Nicolas.Gillis@uclouvain.be
Samuel Gissot
Royal Observatory of Belgium
sgissot@oma.be
Bart Goossens
Ghent University
bart.goossens@telin.ugent.be
Wojciech Gradkowski
UCL
wgradkowski@gmail.com
Hans Hallez
Ghent University
Hans.Hallez@ugent.be
Jan Haskovec
Austrian Academy of Sciences
jan.haskovec@ricam.oeaw.ac.at
Rob Heylen
University of Antwerp
rob.heylen@ua.ac.be
Daan Huybrechs
KULeuven
daan.huybrechs@cs.kuleuven.be
Marian-Daniel Iordache
University of Extremadura
mariandaniel.iordache@yahoo.fr
Laurent Jacques
University of Louvain
laurent.jacques@uclouvain.be
38
Maarten Jansen
K.U.Leuven
maarten.jansen@wis.kuleuven.be
Don Johnson
Rice University
dhj@rice.edu
Felix Krahmer
Universit¨at Bonn
felix.krahmer@hcm.uni-bonn.de
Gitta Kutyniok
Universit¨at Osnabr¨uck
kutyniok@uni-osnabrueck.de
Loic Lecharlier
Universit´e libre de Li`ege Gembloux Agro
Bio-Tech
Loic.Lecharlier@ulb.ac.be
Wen-shin Lee
University of Antwerp
wen-shin.lee@ua.ac.be
Ignace Loris
Vrije Universiteit Brussel
igloris@vub.ac.be
Hiep Luong
Ghent University
hiep.luong@telin.ugent.be
Jianglin MA
Vrije Universiteit Brussel
jianma@vub.ac.be
Benoit Macq
Universit´e Catholique de Louvain
benoit.macq@uclouvain.be
Zahid Mahmood
University of Antwerp
zahid.mahmood@ua.ac.be
David Mary
University of Nice
david.mary@unice.fr
Adrian Munteanu
Vrije Universiteit Brussel
acmuntea@etro.vub.ac.be
Vahid Nassiri
Vrije Universiteit Brussel
vnassiri@vub.ac.be
Jonas Offtermatt
University of Stuttgart
Jonas.Offtermatt@mathematik.uni-
stuttgart.de
Aleksandra Pizurica
Universiteit Gent
Aleksandra.Pizurica@telin.ugent.be
Ljiljana Platisa
UGent
ljiljana.platisa@telin.ugent.be
Javier Portilla
Instituto de Optica, CSIC
portilla@io.cfmac.csic.es
H´erald RABESON
IFP
herald.rabeson@ifp.fr
Holger Rauhut
University of Bonn
rauhut@hcm.uni-bonn.de
39
Joseph Richards
Carnegie Mellon University
jwrichar@stat.cmu.edu
Baena Gall´e Roberto
University of Barcelona
rbaena@am.ub.es
Eduardo Romero
Universidad Nacional de Colombia
edromero@unal.edu.co
Andrea Rueda-Olarte
Universidad Nacional de Colombia
adruedao@unal.edu.co
Tijana Ruzic
Ghent University
truzic@telin.ugent.be
Khalid Sabri
LATT-CNRS
khalid.sabri@ast.obs-mip.fr
Aswin Sankaranarayanan
Rice University
saswin@gmail.com
Jean Schoentgen
Universit´e Libre de Bruxelles
jschoent@ulb.ac.be
Adalberto Schuck J´unior
Federal University of Rio Grande do Sul
adalberto.schuck@uclouvain.be
Iuliia Shatokhina
Johannes Kepler University
iuliia.shatokhina@indmath.uni-linz.ac.at
Marco Signoretto
K.U. Leuven
marco.signoretto@esat.kuleuven.be
Diana Stoeva
Universit´e catholique de Louvain
stoeva fte@uacg.bg
David Stork
Ricoh Innovations
DAVIDGSTORK@GMAIL.COM
Aimamorn SUVICHAKORN
Universite catholique de Louvain
aimamorn@gmail.com
Maja Temerinac-Ott
University of Freiburg
temerina@informatik.uni-freiburg.de
Stefanie Tenorth
University of Duisburg-Essen
stefanie.tenorth@gmx.de
George Tzagkarakis
University    of    Crete    &    Foundation
for   Research   and   Technology-Hellas
(FO.R.T.H.)
gtzag@csd.uoc.gr
Raf Van de Plas
K.U.Leuven
raf.vandeplas@esat.kuleuven.be
Pierre Vandergheynst
EPFL
pierre.vandergheynst@epfl.ch
40
Dimitri Van De Ville
Ecole  Polytechnique  F´ed´erale  de  Lau-
sanne and University of Geneva
Dimitri.VanDeVille@unige.ch
Sabine Van Huffel
Katholieke Universiteit Leuven
sabine.vanhuffel@esat.kuleuven.be
Cis Verbeeck
cis.verbeeck@oma.be
Royal Observatory of Belgium
Nico Verbeeck
K.U. Leuven
nico.verbeeck@esat.kuleuven.be
Caroline Verhoeven
Vrije Universiteit Brussel
cverhoev@vub.ac.be
Sergey Voronin
Princeton University
svoronin@princeton.edu
Mariya Zhariy
Johannes Kepler University
zhariy@indmath.uni-linz.ac.at
Ping Zhu
Royal Observatory of Belgium
zhuping@oma.be
41
5    Practical information
5.1    Lunch
There is a restaurant on the ground floor (floor 0, not 1) of building R.  Meals are
served from 11h30 to 13h45.
The set menus include soup, a main course and a dessert as well as free tap water.
Next to the set menus guests have a choice of pasta, wok, grill, cold dishes and salads.
Meals are to be paid cash at the cash register (present student card for a significant
discount).
The Cafeteria (on the first floor) serves breakfast, sandwiches and drinks.  You
can find also two snack bars on the campus: Complex and Opinio. Opinio is next to
building D.
Here is a link to this week’s menu: http://www.vub.ac.be/infovoor/
toekomstigestudenten/restaurant/menuetterbeekENG.html
6    Getting around
The workshop is held in room D.007 of building D on the campus of the Vrije Uni-
versiteit Brussel in Brussels (Belgium). Refer to the maps on the following pages for
orientation.
Figure 1: Building D
The main entrance of building D (and build-
ings B, C, E F, G, R as well) is on level +1,
via the so-called “Esplanade” (dark grey on the
campus map on page 43).
It takes about 15 minutes to walk from sub-
way station “P´etillon” or “Delta” to building D.
It takes 9 minutes to walk from the railway sta-
tion “Etterbeek” to building D.
Hotel  “Adagio”  is  located  at  Boulevard
Anspach 20, next to exit 4 of metro station “De
Brouckere” (the station has several exits).
Subway tickets are 1.70e/ride, or 12.30e
for a 10-ride card and can be bought at vending machines in the stations.  Do not
forget to validate your ticket before entering the platform (insert ticket in orange box
and take back).  The metro ride from “de Brouckere” to “P´etillon” takes about 15
minutes.
42
6.1    Campus map

                                



                Metro “P´etillon”
(15min)
Av. des volontaires
Vrijwilligerslaan
Bd. du Triomphe/Triomflaan
Bd. de la Plaine/Pleinlaan
Bd. G. Jacques/G. Jacqueslaan
Railway station “Etterbeek”
(9min)
Workshop
Building D
room D0.07
Restaurant
Metro “Delta”
(15min)
E40-E19
(Leuven-Antwerpen)
E411
(Namur)
E411
(Namur)
car park
Entrance 13
(with bar code)
Entrance 6
(with bar code)
43
6.2    By car

 

 
          

 
     
 


 



     !    ”

#$
%

    !


Important:  If you come to
campus by car, use this bar
code  to  enter  through  en-
trance 6 or 13 (the scanner
is under the intercom). Park-
ing is free.
44
6.3    By public transport
Metro “De Brouckere”
(near hotel “Adagio”)
Metro “Gare Centrale/Centraal Station”
(walk to train station to go to airport)
Metro “P´etillon”
(near VUB)
Metro “Delta”
(near VUB)
45
6.4    Near hotel “Adagio”
Rue du marais
Schildknaapstraat
Rue de l’Ecuyer
Anspachlaan
Bd Anspach
Place de la
monnaie
Rue de l’éveque
Rue des fripiers
Galeries Royales
Bourse
rue au beurre
rue d’Arenberg
Bd de l’impératrice
Putterie
Galeries Royales
Rue neuve
Wolvengracht
Nieuwstraat
Place
de Brouckere
Zandstraat
Railway    station
“Centraal”
“Grand Place/Grote Markt”
La Chaloupe d’Or
City hall
Manneken Pis
Hotel “Adagio”
Comic   Book
Museum movie theatre
Metro
“De brouckere”
46
7    Personal notes
47
Interdisciplinary Workshop
SPARSITY
AND
MODERN MATHEMATICAL METHODS FOR HIGH DIMENSIONAL DATA
April 6–April 10, 2010, Vrije Universiteit Brussel, Belgium
In recent years exciting new developments in mathematics and computer science
have opened up new domains of application for computational mathematics.  These
developments bring new challenges, for which new approaches and new tools must
be developed.  These draw not only from traditional linear-algebra-based numerical
analysis or approximation theory, but also from information theory, graph theory, the
geometry of Banach spaces, probability theory, and more. Often the features, patterns
or structures of interest hidden in the data, are typically concentrated on subspaces
or manifolds of much smaller dimensions.  Even if one has no extra a priori knowl-
edge about which subspace or submanifold might carry the information of interest,
the knowledge that it is of much smaller dimension helps in “digging it out”. Taking
advantage of this underlying sparsity lies at the heart of these new developments. It is
also the central tenet of compressed sensing and is presently seeing intense develop-
ment in inverse problems as well.
This workshop will give young scientists in particular the opportunity to present
their recent results on new mathematical methods for high-dimensional data and their
applications, to a broad audience. In addition, a small number of invited speakers will
present their research field in a more general way.
Topics:
•  sparse techniques in inverse problems and compressed sensing (theory, algo-
rithms, applications, . . . )
•  wavelet-like transforms (shearlets, curvelets, use of non-regular grids, . . . )
•  statistical multi-resolution modeling and restoration of images (with applica-
tions in remote sensing, biomedical imaging, . . . )
•  analysis of multi-spectral data and the study of art
The workshop is part of the activities organized on the occasion of the appointment
of Prof.  Daubechies as International Francqui Professor at the VUB (January-June
2010).

sparse

successive interference cancellation under cdma and compressive sensing!

sic

In this paper, the CS framework is introduced firstly, and then the short-term stability of speech signal and the sparsity in the discrete cosine transform basis of speech signal are analyzed. Secondly, a new distributed speech signal compression and reconstruction framework based on compressed sensing theory is proposed. Via basis pursuit(BP) and orthogonal matching pursuit(OMP), it is demonstrated that the performance of reconstruction is correlated with the number of measurements and the length of frames.

By SUN Lin-hui and YANG Zhen, Nanjing University of Posts and Telecommnunications

Compressive Sensing

Download full publication in pdf format

Download chapters in pdf format:

• Acknowledgments
• Contents
• List of Illustrations
• Foreword
• The growing role of knowledge in the global economy
• United States of America
• Canada
• Latin America
• Brazil
• Cuba
• 7. The CARICOM countries
• European Union
• Southeast Europe
• Turkey
• Russian Federation
• Central Asia
• Arab States
• Sub-Saharan Africa
• South Asia
• Iran
• India
• China
• Japan
• Republic of Korea
• Southeast Asia and Oceania
• Annexes
• Statistical annex

UNESCO Science Report 2010
© UNESCO
UNESCO Science Report 2010
Europe, Japan and the USA (the Triad) may still dominate research and development (R&D) but they are increasingly being challenged by the emerging economies and above all by China. This is just one of the findings of the UNESCO Science Report 2010.

Written by a team of independent experts who are each covering the country or region from which they hail, the UNESCO Science Report 2010 analyses the trends and developments that have shaped scientific research, innovation and higher education over the past five years, including the impact of the current global economic recession, which has hit the Triad harder than either Brazil, China or India. The report depicts an increasingly competitive environment, one in which the flow of information, knowledge, personnel and investment has become a two-way traffic. Both China and India, for instance, are using their newfound economic might to invest in high-tech companies in Europe and elsewhere to acquire technological expertise overnight. Other large emerging economies are also spending more on research and development than before, among them Brazil, Mexico, South Africa and Turkey.

If more countries are participating in science, we are also seeing a shift in global influence. China is a hair’s breadth away from counting more researchers than either the USA or the European Union, for instance, and now publishes more scientific articles than Japan.

Even countries with a lesser scientific capacity are finding that they can acquire, adopt and sometimes even transform existing technology and thereby ‘leapfrog’ over certain costly investments, such as infrastructure like land lines for telephones. Technological progress is allowing these countries to produce more knowledge and participate more actively than before in international networks and research partnerships with countries in both North and South. This trend is fostering a democratization of science worldwide. In turn, science diplomacy is becoming a key instrument of peace-building and sustainable development in international relations.

Taking up from where its predecessor left off in 2005, the UNESCO Science Report 2010 proposes a world tour of the status of science today that should enable ‘science watchers’ everywhere to decipher the trends that are shaping our rapidly changing world.

The following links will be activated once the press embargo is lifted at 11 am (Paris time) on 10 November 2010:

science

POSTDOC POSITION AVAILABLE AT INRIA SOPHIA ANTIPOLIS (FR)

Period : 11 months, Feb-Dec 2011.

————–

See also here about how to apply:

http://www.inria.fr/travailler/mrted/fr/postdoc/details.html?id=PGTFK026203F3VBQB6G68LONZ&LOV5=4508&LOV2=4494&LG=FR&Resultsperpage=20&nPostingID=4591&nPostingTargetID=9499&option=52&sort=DESC&nDepartmentID=19

TITLE:
Peer-to-peer storage (backup) simulation and workload characterization

Position: Post-doctorate
Working site: INRIA Sophia-Antipolis

About INRIA and the position
INRIA is a research institute specialized Information and Communication
Sciences and Technologies (ICST). 3800 people work in Research Centers
in 7 regions of France. The Sophia Antipolis Research Center hosts 500
people, working in 30 research teams and administration. Mascotte is a
joint team between INRIA and the laboratory I3S which itself belongs to
CNRS and University of Nice-Sophia Antipolis (UNS). MASCOTTE’s main
objective is to develop algorithmic methods and tools, with particular
emphasis on the design of telecommunication networks. This involves high
level research in the fields of simulation, algorithms, and discrete
mathematics. Overlay networks and peer-to-peer systems have been studied
in MASCOTTE since 2007. Because of their very large scale, the study of
these systems generates challenging problems both from theoretical and
practical points of views. MASCOTTE adresses both kinds of issues using
a wide palette of tools, ranging from stochastic models theoretical
analysis to simulations and experimentations. In order to share
experience and join efforts with other French research teams working on
similar issues, MASCOTTE has joined the SIMGRID ANR project, whose goal
is to provide simulation tools for large scale distributed systems and
give practical demonstration of their use. The post-doc candidate will
work on this project under the supervision of Olivier Dalle and in
collaboration with other MASCOTTE researchers.

Mission
The work consists in contributing to the SIMGRID project at user level:
rather than working at the core of the SIMGRID simulator architecture,
the contribution is expected help and illustrate the use of the
simulator. This contribution will be two-fold: first the design and
implementation of a low-level mechanism to capture actual workloads and
replay them efficiently in the SIMGRID simulator; and second, use the
SIMGRID simulator to run some practical performance evaluation study of
P2P storage systems.

Work description
The work will be divided in two tasks:
1. Application Workload Characterization. The goal of this task is to
capture the workload through a set of defined events. Some of them (such
as send and receive) are shared between all applications, but they are
very low-level. Higher level events have to be specific to the
application. For example, in a P2P DHT, join or lookup are classic
events while submit is often seen in a batch scheduler context. This
task has two goals. First, we aim at implementing an instrumentation
tool able to capture the low-level events of any application (using
system-level solutions such as ld preload), and record them accordingly
to a generic event format. Then, we want to provide a solution to
generate the simulation code corresponding to a given event log. We do
not plan on providing a tool to capture high-level traces since they are
too application-specific for us to devise a generic tool.
2. Peer-to-peer backups: Simulation environments for large scale
distributed applications, such as peer-to-peer video on-demand systems
or generic peer-to-peer storage systems, are generally limited to the
estimation of metrics such as the number of messages exchanged between
the different peers and do not consider timing issues. In the particular
case of peer-to-peer backup, being able to estimate the time needed to
load or store a file chunk is crucial. We expect the use of such a tool
to provide a better understanding of the behavior of a working backup
system, and in particular, to compute some parameters that impact the
performance of the system and are hard to guess from standard
simulations (sizes of volumes, sizes of chunks, failure detection
delays) where network characteristics are taken into account enough.

Profile
The candidate must hold a PhD in Computer Sciences with significant
experience in network performance evaluation and simulation techniques.
The candidate must also be skilled for programming in C language, be
interested in producing high quality research publications and be ready
to participate regularly the ANR SIMGRID project meetings. A previous
experience of the SIMGRID simulator is appreciated but not mandatory.

_______________________________________________
Tccc mailing list
Tccc@lists.cs.columbia.edu

https://lists.cs.columbia.edu/cucslists/listinfo/tccc

position

Call for Papers

********************************************************************************************
IEEE Transactions on Vehicular Technology
Special Section: Telematics Advances for Vehicular Communication Networks
********************************************************************************************

Wireless communication for intelligent transportation systems (ITSs) is
a promising technology to improve driving safety, reduce traffic
congestion and support information services in vehicles. A new era of
vehicular technology that includes vehicle-to-vehicle (V2V) and
vehicle-to-infrastructure (V2I) communications is approaching. During
recent ITS development, transportation telematics techniques have
exhibited much progress, e.g., interaction between automobiles and the
infrastructure for delivering services such as road-side assistance,
automatic crash notification, concierge assistance and vehicle condition
reports. A number of IEEE 802.11p-like equipment prototypes have been
built, and several technical reports based on field trials have
demonstrated the lack of cutting-edge techniques to improve system
performance. Technology and applications for ITSs and telematics design
are rapidly emerging, and there is a critical need to bring together
professional researchers, engineers, academia, industry, standard
committees, the private and public sectors to exchange new ideas. This
special section aims to spur research progress by serving as a forum in
which both academia and industry can share experiences and report
original work regarding all aspects of vehicular communication, e.g.,
vehicular ad hoc networks (VANETs), information dissemination, road
safety, ITS and emergency services. Our primary goal is to promote
meaningful research in the cross-layered design of architectures,
algorithms and applications for inter-vehicle communication
environments. This special section will also address numerous
significant standardization efforts (IEEE 802.11p, p1609, TIA TR48,
etc.) and some alternatives or improved systems.

Topics of Interest:

* Data-collection, organization and dissemination methods:
- Floating vehicles
- Traffic and flow modelling and analysis
- Remote service provisioning and over-the-air upgrading technology
- Data replication, caching and pre-fetching protocols
* V2V and V2I communications:
- Network protocols including MAC, routing, addressing, multicast, TCP
protocols and end-to-end quality of service, resource management,
security and privacy
- Design with multiple wireless data links (802.11p, WiMAX, WiFi, cell
phone, GPS)
- Mobility or handover technology
- System-level, board-level and chip-level electronics
- PHY issues: channel measurement, channel modelling, channel
estimation, antenna arrangement, pilot arrangement, etc.
* New ITS/Telematics applications:
- Safety and driver-assistance applications
- Congestion control by cooperative data analysis
- Reduction of fuel consumption and greenhouse gas emission
* Ongoing ITS/Telematics activities:
- Results from large-scale experimental systems, test beds and field
trials
- Hardware implementation and infrastructure deployment
- Deployment strategies and predictions
- Standardization and development of VANETs: efforts and problems on
802.11p WAVE, 802.11s MESH, DSRC, etc.

Paper Submission:

Authors should follow the IEEE TVT manuscript format and submission
procedure which can be found at the IEEE TVT home page
http://transactions.vtsociety.org/ under Information for Authors.
Prospective authors should submit a PDF version of their complete
manuscript via the journal online paper submission system at

http://mc.manuscriptcentral.com/tvt-ieee

Important Dates:

* Manuscript Submission Due: January 31, 2011
* Notification Letter Sent: April 15, 2011
* Revision Due: May 31, 2011
* Acceptance Notification: July 31, 2011
* Final Manuscript Due: August 31, 2011

Guest Editors:

- Jia-Chin Lin, Department of Communication Engineering, National
Central University, Taiwan
- Christoph Mecklenbrauker, Institute of Communications and
Radio-Frequency Engineering, Vienna University of Technology, Austria
- Alexey Vinel, Tampere University of Technology, Finland
- Tao Zhang, Emerging Technologies and Services Research, Telcordia
Technologies, Inc., USA
- Spyridon Vassilaras, Athens Information Technology Center for Research
and Graduate Education, Greece
- Kuen-Rong Lo, Telecommunication Laboratories, Chunghwa Telecom Co.,
Ltd., Taiwan

A pdf version of this CFP can be found at:

http://bbcrlab-pc9.bbcrlabpcnet.uwaterloo.ca/tvt/vtjournal/CFP-VANET-2010.pdf

_______________________________________________
Tccc mailing list
Tccc@lists.cs.columbia.edu

https://lists.cs.columbia.edu/cucslists/listinfo/tccc

cfp

Hot-ICE ‘11 Call for Papers
————————————

Workshop on Hot Topics in Management of Internet, Cloud, and Enterprise Networks and Services
March 29, 2011
Boston, MA

http://www.usenix.org/events/hotice11/

Sponsored by USENIX, the Advanced Computing Systems Association

Hot-ICE ‘11 will be co-located with the 8th USENIX Symposium on Networked Systems Design and Implementation (NSDI ‘11), which will take place March 30–April 1, 2011.

Important Dates:
Paper registration due: December 9, 2010, 11:59 p.m. PST
Paper submissions due: December 16, 2010, 11:59 p.m. PST
Notification of acceptance: February 15, 2011
Final papers due: February 28, 2011

Overview

The first Hot-ICE workshop seeks to bring together researchers and practitioners working on network and service management in the Internet, cloud, and enterprise domains.

The scope of Hot-ICE includes all aspects of network and service management. This includes traditional network management concerns, management of network services and/or services enabled by networks, management of clean-slate network architectures, and clean-slate designs of management architectures. We seek new ideas and experimental or operational insights that help make Internet, cloud, and enterprise networks and services more secure, more systematically or automatically configurable, more scalable, and able to achieve more predictable performance, better accountability, greater fault tolerance, and faster fault recovery.

Topics

Topics of specific interest include but are not restricted to:

- Novel network and service management systems
- Greenfield network management architectures, and management of new network architectures
- The use of data-mining techniques in network and service management
- Network and service management aspects of existing and emerging network architectures (e.g., data-center networks, cloud architectures, data-centric architectures, software-defined architectures, and mobile networks)
- Management approaches that involve cross-domain and cross-layer techniques
- Management techniques and tools for the verification, synthesis, diagnosis, and evaluation of network operations and policies

We invite short position papers or work-in-progress reports. Hot-ICE will particularly favor interesting and new ideas and early results that lead to well-founded position papers, i.e., papers that illustrate a firm understanding of the problem and can position the contribution in the broader context of related work. Once fully developed and evaluated, we envision that work presented at Hot-ICE will be published at relevant, high-quality conferences.

Papers will be selected primarily based on technical merit and originality, with additional consideration given to their potential to generate discussion at the workshop.

Hot-ICE evolved from earlier Internet Network Management (INM) workshops, which more recently combined with the Workshop on Research on Enterprise Networking (WREN). As such, Hot-ICE, while a new workshop, is serving an established community, but with a broader scope, in recognition of the evolving concerns of the community. Hot-ICE is modeled after other “Hot”-style workshops, seeking to provide a venue for discussing innovative ideas and early results in network and service management that have the potential to significantly influence the community.

Please contact the program co-chairs if you have questions concerning the relevance of your topic of interest.

Program Co-Chairs:

Anees Shaikh, IBM Research
Kobus Van der Merwe, AT&T Labs—Research

Steering Committee:

Ehab Al-Shaer, University of North Carolina, Charlotte
Albert Greenberg, Microsoft Research
Chuck Kalmanek, AT&T Labs—Research
David Maltz, Microsoft Research
Jeff Mogul, HP Labs
Tze Sing Eugene Ng, Rice University
Geoffrey Xie, Naval Postgraduate School

_______________________________________________
Tccc mailing list
Tccc@lists.cs.columbia.edu

https://lists.cs.columbia.edu/cucslists/listinfo/tccc

.

cfp

***  CALL FOR PAPERS ****
IQ2S 2011
The Third International Workshop on Information Quality
and Quality of Service for Pervasive Computing
www.iq2s.org

in Conjunction with IEEE PERCOM 2011
Seattle, USA, March 21, 2011

IMPORTANT DATES

Paper Submission:             October 31, 2010 (New Date)
Acceptance Notification:       January 7, 2011 (New Date)
Camera-Ready Due:               January 28, 2011 (New Date)

The Program Committee for IQ2S 2011 is soliciting original papers
addressing
both theoretical and practical aspects of QoI and QoS provisioning in
pervasive
computing. The objective of this workshop is to provide a forum to exchange
ideas, present results, share experience, and enhance collaborations among
researchers, professionals, and application developers in various
aspects of QoI
and QoS in wireless sensor networks for pervasive computing.  Papers
describing
experiences on real prototype implementations are particularly welcome.

Topics of interest addressing the challenging joint aspects of QoI and QoS
include:
* Joint QoI- & QoS-driven system design and architectural principles
* Network services (time sync, QoS) for target/event detection,
localization,
tracking and classification
* QoI-aware wireless sensor networking
* Energy-efficient data fusion, sensor fault analysis, sensor data cleansing
* QoS for task mapping and scheduling
* Coordinated QoS for cross-layer, cross-application, and
cross-node integration (including QoI-QoS integration)
* Query optimization for event processing in pervasive environments
* Data and query models for QoI-aware event processing
* Adaptive QoI and QoS under dynamic environments
* Trust, security, privacy, and data provenance issues in QoI and QoS
* QoI characterization, representation, performance metrics, and evaluation
* QoI and QoS for emerging pervasive computing applications
* Quality of Experience (QoE) issues for pervasive applications
* Value of information and quality of action for sensor/actuator networks
* Prototype test-bed design, implementation, and field trials

Further details, including the list of TPC members and submission
instructions,
are available via the Web site: http://www.iq2s.org
Regards,
Archan Misra & Kai-Uwe Sattler
iq2s2011-chairs at tu-ilmenau.de<mailto:iq2s2011-chairs at tu-ilmenau.de>
IQ2S 2011 TPC Co-chairs

_______________________________________________
Tccc mailing list
Tccc@lists.cs.columbia.edu

https://lists.cs.columbia.edu/cucslists/listinfo/tccc

cfp