Abstract: In this talk, we show two possible applications of trees in deep learning, one in the area of classification and the other in clustering. In the first part of the talk, we construct a (non-binary) tree decision model which simultaneously generalizes binary trees and classical softmax, and show how to efficiently implement its gradient training (this part of the talk will be motivated by https://arxiv.org/pdf/2107.13214.pdf WACV 2023). In the second part we present clustering in deep networks with the use of application of contrastive learning to binary trees (results based on https://openreview.net/pdf?id=Hv57u3WQ0WZ).

Speaker: Jacek Tabor, https://scholar.google.com/citations?user=zSKYziUAAAAJ&hl=enm, holds the professor position at the Jagiellonian University, Kraków, Poland. He is the leader of the research deep learning group GMUM https://gmum.net/ at the Jagiellonian University, Cracow, Poland. The scientific interests of Jacek Tabor concern the classification of images, generative models (VAE-like, GANs, flow models, etc), and meta-learning. He is the author of around 100 papers in international journals and conferences and publishes actively in A and A* conferences like NeuIPS, ICML, ECCV, KDD, etc.  Jacek Tabor has obtained many research grants, in particular he is the leader of grant FNP TEAM NET: Biologically inspired neural networks http://bionn.matinf.uj.edu.pl/ (the total amount of around 4mln euro).

Place of Seminar:  Otaniemi, T5 (zoom)