Generative Modeling with Diffusion

TL;DR

This paper reviews diffusion models as a generative technique, detailing their processes, training algorithms, and exploring their application in enhancing classifier performance on imbalanced datasets.

Contribution

It provides a formal overview of diffusion models, including algorithms for training and generation, and investigates their use in imbalanced data classification.

Findings

Diffusion models can generate high-quality samples.

Applying diffusion models can improve classifier accuracy on imbalanced data.

The paper formalizes the noising and denoising processes in diffusion models.

Abstract

We provide an overview of the diffusion model as a method to generate new samples. Generative models have been recently adopted for tasks such as art generation (Stable Diffusion, Dall-E) and text generation (ChatGPT). Diffusion models in particular apply noise to sample data and then "reverse" this noising process to generate new samples. We will formally define these noising and denoising processes, then present algorithms to train and generate with a diffusion model. Afterward, we will explore a potential application of diffusion models in improving classifier performance on imbalanced data.