Xianzhi’s Tech Notebook

Here’s a link to my portfolio website: https://xianzhiwang1.github.io/.

Denoising Diffusion Probabilistic Models from Scratch

A PyTorch implementation and walkthrough of Ho et al. (2020), trained on MNIST

deep learning

generative models

diffusion

pytorch

This report documents an end-to-end PyTorch implementation of the Denoising Diffusion Probabilistic Model (DDPM) introduced by Ho, Jain, and Abbeel (2020). After a self-contained derivation of the forward and reverse diffusion processes, we walk through every component of the implementation: the linear \(\beta\) variance schedule, the closed-form forward marginal \(q(x_t \mid x_0)\), the noise-prediction reparameterization \(\epsilon_\theta\), and the time-conditional U-Net built from residual blocks, self-attention, and sinusoidal time embeddings. We train the model on the MNIST handwritten-digit dataset and present qualitative samples produced by the trained network.

Nov 15, 2025

Xianzhi Wang

Classifying Palmer Penguins

machine learning

classification

supervised learning

palmer penguins

My blog post on Palmer Penguins classification

May 3, 2023

Xianzhi Wang

Machine Learning Final Project using Economics Historical Data

machine learning

classification

logistic regression

economics

data analysis

Final Project on prediction of Incorporation using replication data file on Russian Historical Factory level data from Professor Gregg, Middlebury College, Vermont

May 2, 2023

Prateek and Xianzhi