projects

A growing collection of my projects, mostly coursework.

project thumbnail

Empirically Testing the Boundaries of the Frequency Principle

Xu (2019) and Rahaman (2019) propose that deep neural networks often fit target functions from low to high frequencies during the training process. To what extent does this principle hold?
project thumbnail

Gibbs Sampling and Variational Inference for HDP-HMM

PoS tagging is essential for NLP tasks, yet supervised methods depend heavily on annotated data. Unsupervised methods like Hierarchical Dirichlet Process Hidden Markov Models (HDP-HMM) offer a promising alternative. This study provides a detailed derivation of a Gibbs sampling and a variational inference algorithm for HDP-HMM and implements the code.