PL condition is an alternative assumption of convexity in the optimization analysis of deep neural networks, which might help explaining the landscape of the objective function for DNNs.

The expressivity or approximation theory of neural networks is a much more well-understood topic than the optimization and generalization of NNs. The survey paper Gühring et al. (2020) has a comprehensive review.

In this notebook, we explore some popular matrix-decomposition based recommeder system algorithms on Movielens100K dataset. This is largely based on the slides of STAT8056, Spring 2021, which is taught by Prof. Xiaotong Shen.

This blog records how I install and configure my Ubuntu 20.04 . All the codes should be run in a terminal.

This blog records the way how I build and publish my first Python package, SubsetPrivacy (not released yet).