JISUNG
HWANG

Understanding the Role of Noise in Neural Networks

I am a first-year Ph.D. student in the Visual AI Group at KAIST, advised by Prof. Minhyuk Sung. I study how Gaussian noise and whitening shape the behavior of generative models, and more broadly what enforcing or exploiting noise structure can offer neural networks at training and inference time.

This theme has driven my work on Gaussianity regularization and gradient preconditioning via white noise projection, and now draws me toward the Muon optimizer, where I see deeper connections to whitening waiting to be uncovered.

I previously worked on AI for bioscience and 3D computer graphics. Recently, I have become interested in efficient CUDA implementations, because making ideas practical at scale matters as much to me as the ideas themselves.