Byungkon Kang's research focuses on developing and modeling algorithms for a variety of machine learning problems.

In particular, his current domain of interest is learning representations (in the form of vector embedding) for various objects such as natural languages, images, and bio-medical data.

Another branch of research is about deep generative models, where he studies and exploits neural network architectures that efficiently model joint probability distributions of data.

In addition to these current topics, his past research interests include reinforcement learning and stochastic planning.