Ofir Lindenbaum Guest Lecture From BIU – Learning Club 19.6.22

The recording of Ofir’s talk and slides is available here.

###### Meeting Info ######

Title:
Deep learning for tabular biomedical data

Abstract:
Biomedical datasets are often low-sample-size or high-dimensional. Practitioners in this domain prefer linear or tree-based models over neural networks since the latter are harder to interpret and tend to overfit when applied to tabular datasets. I will describe our recently proposed frameworks that address these shortcomings in this talk. First, I will show how we can incorporate Gaussian Stochastic Gates (STG) into the input layer of a NN to remove irrelevant (i.e., noisy and information-poor) features. Under a linear data model, I will show theoretically and empirically that the STG can successfully recover the set of informative features. Next, I will show how to exploit redundancy to extract information that is shared across unlabeled multimodal biomedical datasets. Towards this goal, we propose L0-Deep Canonical Correlation Analysis, a method for learning correlated representations based on sparse subsets of variables from two observed modalities. Finally,  I will show how locally sparse NN can reduce model overfitting in low-sample size data while remaining highly interpretable.

References:
– J. Yang*, O. Lindenbaum*, Y. Kluger, Locally Sparse Neural Networks for Tabular Biomedical Data, ICML, 2022
– O. Lindenbaum, M. Salhov, A. Averbuch, Y. Kluger, L0-based Sparse Canonical Correlation Analysis, ICLR 2022‏
– Y.Yamada*, O. Lindenbaum*, S. Negahban, Y. Kluger., “Feature selection using Stochastic Gates.” International Conference on Machine Learning (ICML), 2020.
– Jana, S., Li, H., Yamada, Y.,  Lindenbaum, O. Support Recovery with Stochastic Gates: Theory and Application for Linear Models. arXiv (2022)

Speaker’s Short bio:
Ofir Lindenbaum is a senior lecturer in the faculty of Engineering at Bar Ilan University. Ofir obtained his Ph.D. and M.Sc. from Tel Aviv University and B.Sc. in Electrical Engineering and Physics (both summa cum laude) from the Technion. Following his Ph.D., he served as a Gibbs assistant professor at Yale University. His research is focused on the theory and practice of machine learning. His main goal is to enable the practical use of machine learning algorithms for scientific discovery.

##### Connection Details #####
Join Zoom Meeting
https://us02web.zoom.us/j/7359565559

Meeting ID: 735 956 5559
One tap mobile
+13017158592,,7359565559# US (Washington DC)
+13126266799,,7359565559# US (Chicago)

Dial by your location
+1 301 715 8592 US (Washington DC)
+1 312 626 6799 US (Chicago)
+1 346 248 7799 US (Houston)
+1 646 558 8656 US (New York)
+1 669 900 9128 US (San Jose)
+1 253 215 8782 US (Tacoma)
Meeting ID: 735 956 5559
Find your local number: https://us02web.zoom.us/u/kbjcWyWAWq