A Johnson–Lindenstrauss framework for randomly initialized CNNs
Information Theory & Computer Communications
The information velocity of packet-erasure links
About the Lab
The research carried in the lab is mathematically focused and spans the broad areas of Information Theory, Communications, Control Theory, Signal Processing, Statistics, and Learning Theory.
In particular, we study various aspects of causality: measures of causality, efficacy, and dependence, as well as distilling them via learning; applications with causality constraints: low-delay communications, real-time signal processing, and control.
Some Recent Works
Learning Theory
A Johnson–Lindenstrauss framework for randomly initialized CNNs
with Ido Nachum, Jan Hązła, and Michael Gastpar (all @EPFL)
Information Theory & Computer Communications
The information velocity of packet-erasure links
with Elad Domanovitz and Tal Philosof (Samsung Research)
Communications & Signal Processing
Energy-limited joint source–channel coding via analog pulse position modulation
with Omri Lev
Control Theory & Learning Theory
Learning-based attacks in cyber-physical systems
with Mohammad Javad Khojasteh, Massimo Franceschetti, and Tara Javidi (all @UCSD)
Statistical Signal Processing
Monotonicty of the trace–inverse of covariance submatrices and two-sided prediction
with Arie Yeredor and Ram Zamir
Anatoly Khina
Principal investigator