Improving inference algorithms for diffusion models

, ,

Professor: Thuy-Duong (June) Vuong

Description: Diffusion models struggle with generating long sequences, as errors accumulate over iterative steps, degrading temporal or structural consistency. Current approaches mitigate this by generating short sequences and chaining them, but this introduces compounding errors. Addressing this issue is critical for applications like video generation, weather forecasting, and physical simulations.

Preferred Qualifications: Very good understanding of mathematics, probability theory, and algorithm designs. Programming skills in Python.

Contact Us