CFP ICML workshop 2012: Sparsity, Dictionaries and Projections in Machine Learning and Signal Processing


Sparsity, Dictionaries and Projections in Machine Learning and Signal
ICML 2012 Workshop, Edinburgh, Scotland
30 June or 1 July, 2012

Submission Deadline: Monday, May 7, 2012 (5:00 pm PDT)

Sparse representations are today key in many fields of applied
mathematics faced with data, from signal processing to machine
learning and statistics.

Historically, several communities proposed various approaches to
sparse coding: on the one hand the use of carefully crafted
dictionaries, like wavelets, forming bases of functional spaces with
good approximation properties over a class of signals, but constructed
without data; on the other hand, the use of representations derived
directly from data via either algebraic formulations like sparse
matrix factorization, or via probabilistic formulations, based on the
introduction of latent variables, such as, e.g., independent component
analysis or latent Dirichlet allocation.

The introduction of sparsity and/or structure in matrix factorization
scheme, which where previously used for dimensionality reduction,
induced major shifts in several existing paradigms and led to
significant breakthroughs, which have demonstrated the ability of
sparse models to provide concise descriptions of certain high
dimensional data through low-dimensional projections, together with
algorithms of provable performance and bounded complexity. Compressed
sensing (and more generally the clever use of random low-dimensional
projections), dictionary learning, and non-parametric topic models,
are just a few of the rapidly emerging paradigms in this area at the
confluence of signal processing and machine learning.

While sparse models and random low-dimensional projections are already
at the heart of several success stories in signal processing and
machine learning, their full potential is yet to be achieved and calls
for further understanding. The goal of the workshop is to confront the
various point of views and foster exchanges of ideas between the
signal processing, statistics, machine learning and applied
mathematics communities.

We encourage submissions exploring various aspects of learning sparse
models and/or latent representations, in the form of new algorithms,
theoretical advances and/or empirical results. Some specific areas of
interest include structured matrix factorization algorithms, Bayesian
models for latent variable representations, analysis of random
dictionaries versus learned dictionaries, novel applications of
dictionary learning or relationships to compressed sensing.

Submission Guidelines

Submissions should be written as extended abstracts, no longer than 4
pages in the ICML latex style. Style files and formatting instructions
can be found at
Submissions must be in PDF format. Authors’ names and affiliations
should be included, as the review process will not be double blind.
The extended abstract may be accompanied by an unlimited appendix and
other supplementary material, with the understanding that anything
beyond 4 pages may be ignored by reviewers.

Please send your PDF submission by email to
with the words “ICML workshop submission” in the title by 5:00 pm PDT
on Monday, May 7. Notifications will be given on or before 21 May.
Please include these words “ICML workshop” in the title of mails you
are sending about the workshop or workshop submissions. Work that is
pending review, was recently published or was presented elsewhere will
be considered, provided that the extended abstract mentions this
explicitly. Finally, note that there will be no official proceedings
from this workshop.

Invited Speakers (to be confirmed)

Pierre Comon (CNRS / University Nice Sophia Antipolis)
Julien Mairal (University of California at Berkeley)
Matthias Seeger (Ecole Polytechnique Federale de Lausanne)
Daniel Vainsencher (Technion – Israel Institute of Technology)


Michael Davies (Edinburgh), Rémi Gribonval (INRIA), Rodolphe Jenatton
(CNRS) and Guillaume Obozinski (INRIA)