Speaker: Meir Feder (Tel-Aviv University)
Title: Universal Learning for Individual Data
Abstract:
Universal learning is considered from an information theoretic point of view
following the universal prediction approach pursued in the 90's by F&Merhav. Interestingly, the extension to learning is
not straight-forward. In previous works we considered on-line learning and
supervised learning in a stochastic setting. Yet, the most challenging case is
batch learning where prediction is done on a test sample once the entire
training data is observed, in the individual setting where the features and
labels, both training and test, are specific individual quantities. This work
provides schemes that for any individual data compete with a "genie"
(or reference) that knows the true test label. It suggests design criteria and
derive the corresponding universal learning schemes. The main proposed scheme
is termed Predictive Normalized Maximum Likelihood (pNML).
As demonstrated, pNML learning and its variations provide
robust, "stable" learning solutions that outperforms the current
leading approach based on Empirical Risk Minimization (ERM). Furthermore, the pNML construction provides a pointwise indication for the
learnability that measures the uncertainty in learning the specific test
challenge with the given training examples - thus the learner knows when it
does not know. The improved performance of the pNML,
the induced learnability measure and its utilization are demonstrated in
several learning problems including deep neural networks models.
Joint work with Yaniv Fogel and Koby Bibas
Biography:
Professor Meir Feder received the Sc.D. degree from the Massachusetts Institute
of Technology (MIT) and Woods Hole Oceanographic Institution (WHOI) in 1987. He
is now a Chaired Professor and Head of the School of Electrical Engineering,
Tel-Aviv University. He was a visiting Professor at MIT and had visiting
appointments at Bell laboratories and Scripps Institute of Oceanography.
While serving in the Israeli defense Forces, he was awarded the 1978 “creative
mind” award of the chief Intelligence officer. He received the 1993 best paper
award of the Information Theory Society. He was the recipient of the 1994 prize
of Tel-Aviv University for excellent young scientists, the 1994 award of the
Electronic Industry in Israel (awarded by the president of Israel), and the
1995 research prize in applied electronics awarded by Ben-Gurion University. He
is a Fellow of the IEEE for his contribution to universal data prediction and
universal compression.
In parallel to his academic career he was closely involved in the high-tech
industry with numerous companies. In the early 90's he worked with Intel on the
MMX architecture and designed efficient multimedia algorithms for it. In 1998
he co-founded Peach Networks, a provider of server-based interactive TV system
via the cable network, which was acquired in 2000 by Microsoft. He then
co-founded Bandwiz, to provide massive content
delivery systems via "rateless codes". He
is currently the Chief Scientist of Amimon, a company
he co-founded in 2004. Amimon's
"video-modem" technology is the basis of the WHDI (Wireless Home
Digital Interface) standard, initiated by the leading Consumer Electronics
companies, for wireless high-definition A/V connectivity at the home.