Discrete Binomial Diffusion on Network

Abstract

Quals talk:

Learning to generate graphs is challenging as a graph is a set of pairwise connected, unordered nodes encoding complex combinatorial structures. Recently, several works have proposed graph generative models based denoising diffusion models. While graphs are discrete data, most of the currently diffusing models are not able to generate directly from discrete initial state. Besides, current models are not able to generate large network due to the high dimensionality. In this work, we study the case of binomial diffusion, in which the converging distribution results in Erdos-Renyi random graph with p=0. We analyze the behavior of such diffusion process, which can be used to improved the capability of the denoising model and the computation efficiency the generative process.

Please join meeting in Cummings 270.

Zoom is not available for this event; please disregard dial-in passcode included in email.