Representation Learning

Learning robust but discriminative representations is a key ingredient to successfully apply deep learning models in real-world problems. Our goal is to let our models learn abstract representations that can carry good semantic and structural meanings. We are working to apply these representation learning methods to a variety of real-world tasks, including reinforcement learning, computer vision.

Reinforcement Learning
Although deep reinforcement learning(RL) has achieved notable successes in a range of tasks, the high variability of performance makes it difficult to apply deep RL in the real world. DAVIAN Lab. attempt to make RL algorithms more reliable by applying self-supervised representation learning. We are also doing research to increase the interpretability of RL models using finite state machines.