There have been various applications of stochastic differential equation (SDE) theories in machine learning research. In this talk, we review how SDE helps to understand the stochastic behaviors of SGD. In addition to being a tool in the analysis of ML algorithms, SDE itself became a part of algorithms in generative modeling recently. We will briefly review the score-based diffusion models and discuss a new result of 'Soft Truncation'. This talk is based on a work with Dongjun Kim and Il-Chul Moon.
There have been various applications of stochastic differential equation (SDE) theories in machine learning research. In this talk, we review how SDE helps to understand the stochastic behaviors of SGD. In addition to being a tool in the analysis of ML algorithms, SDE itself became a part of algorithms in generative modeling recently. We will briefly review the score-based diffusion models and discuss a new result of 'Soft Truncation'. This talk is based on a work with Dongjun Kim and Il-Chul Moon.