University of Illinois at Chicago and Research Group on AI, University of Szeged
György Turán received his Ph.D. at the Jozsef Attila University in Szeged in 1982. He was at a visiting position at the Institut fur Operations Research of the University of Bonn in 1983-84. Currently he is a Professor at the Department of Mathematics, Statistics and Computer Science at the University of Illinois at Chicago, and a Senior Research Fellow at the MTA-SZTE Research Group on AI of the ELRN at the University of Szeged. His current main interest is interpretability in machine learning. Previously he worked in complexity theory, computational learning theory, commonsense reasoning and combinatorial and logic problems related to these topics.
Interpretability of deep-learned error-correcting codes
Error-correcting codes have been studied since Shannon's work more than 70 years ago, and many families of good codes have been designed using algebraic and combinatorial methods. Recently, codes have been constructed using deep learning. Are these codes similar to traditional ones, or do they provide new types of codes? This question is important from the practical point of view, and it is also interesting as a case study of incorporating deep learning into scientific research. We present approaches to interpreting a particular deep-learned code using techniques such as discrete optimization, influence, property testing and Fourier analysis.
Joint work with N. Devroye, N. Mohammadi, A. Mulgund, H. Naik, R. Shekhar, Y. Wei and M. Zefran.