3:59 p.m.
Dear colleagues,
I'm pleased to announce a talk by Alex Bronstein.
WHEN: Wed., Sep. 19th; 11:30-12:00
WHERE: Room 222, Galleria 1
TITLE: Learning efficient parsimonious models
SPEAKER: Alex Bronstein, Tel Aviv University
ABSTRACT:
Parsimony, preferring a simple explanation to a more complex one, is
probably one of the most intuitive heuristic principles widely adopted
in the modeling of nature. The past two decades of research have shown
the power of parsimonious representation in a vast variety of
applications from diverse domains of science, especially in signal and
image processing, computer vision, and machine learning. A common form
of parsimony is sparsity, postulating that data can be represented by
a small number of non-zero coefficients in an appropriate dictionary.
This model is satisfied by many classes of natural signals and can be
efficiently pursued through convex, greedy, and other optimization
methods. Other manifestations of parsimony successfully capturing more
intricate data structures are various forms of structured sparsity and
low rank matrices and tensors.
Existing parsimonious modelling approaches are model-centric,
following the same pattern: First, an objective comprising a fitting
term and parsimony-promoting penalty terms is constructed; next, an
iterative optimization algorithm is applied to minimize the objective,
pursuing either the parsimonious representation of the data in a given
dictionary, or the dictionary itself. Despite the steady improvement
of iterative optimization tools, their inherently sequential structure
and data-dependent complexity and latency constitute a major
limitation in many applications requiring real-time performance or
involving large-scale data. Also, representations obtained via
optimization are hard to incorporate into a higher-level optimization
problem, practically restricting existing parsimonious models to
unsupervised regimes. Consequently, these models are typically
generative rather than discriminative, rendering difficult several
important applications such as similarity learning for large-scale
information retrieval.
We present a novel approach, in which the emphasis is moved from the
model to the pursuit algorithm, and develop a process-centric view of
parsimonious modeling. We show a family of fixed-complexity processes
derived from proximal descent algorithms and tuned by learning,
capable of outperforming traditional parsimonious modeling approaches
based on optimization. We demonstrate several applications, in which
the new framework achieves state-of-the-art result, including
single-channel audio separation, denoising, and speaker
identification, as well as background subtraction in video.
BIO:
Alex Bronstein received the B.Sc. and M.Sc. with honors from the Department of Electrical
Engineering, and Ph.D. from the Department of Computer Science, Technion, and in currently
with
the School of Electrical Engineering at Tel Aviv University where he is the founder and
head of the
3D Imaging and Vision laboratory, and the head of the Digital Signal Processing
laboratory.
His main research interests are theoretical and computational methods in
computer vision, pattern recognition, shape analysis, and machine learning.
He is a coauthor of the first book systematically treating computational analysis of
deformable shapes,
and chaired the first IEEE workshops on the topic.
Alex Bronstein is the alumnus of the Technion Excellence Program and the Academy of
Achievement.
His research was recognized by numerous awards, including the Hershel Rich Technion
Innovation award,
the Gensler counter-terrorism prize, the Adams Fellowship, and the Krill prize by the Wolf
Foundation.
Dr. Bronstein held visiting appointments in several universities including Stanford.
In addition to his academic activities, he was a co-founder of the Silicon Valley startup
Novafora, where he served as vice president of video technology, and one of the principal
technologists and inventors behind the 3D acquisition technology developed by the Israeli
startup Invision and acquired by Intel in 2012.
=============
Jonathan Masci