452SM - DEEP LEARNING 2020

References different than books

Methods for analyzing and comparing DNN representations

Pairwise comparison of hidden representations produced by DNNs is a hard and daunting task. Besides the very large dimensionality in which these representations live, one has to take into consideration invariances that common similarity metrics (e.g. cosine similarity) lack. We will be presenting a handful of such metrics (e.g. CCA, CKA…) and the ideas behind them.

In the second part of the lecture, we will show the implementation of these techniques in PyTorch, showing also different ways to extract the hidden representations from DNNs.

Slides for today's lecture on Automatic Differentiation by prof. Luca Manzoni.

Slides (in pptx and pdf format) for the introductory lecture to Part II.

Slides for the Keynote speech by D. Doimo and A. Glielmo.

Colab link: https://colab.research.google.com/drive/1fTxE0GWb5BobZhL3j6G6Ra5hBj__c9X-?usp=sharing