Reinforcement Learning

Our reinforcement learning research (RL) is aimed at developing novel algorithms, techniques, and methodologies.  This entails exploring innovative approaches to address fundamental challenges such as exploration-exploitation trade-offs, sample efficiency, and generalization. This involves investigating techniques to enable agents to apply learned knowledge to new, unseen tasks or domains, promoting the development of more flexible and adaptive RL systems. The encompass meta-learning, hierarchical RL, and transfer learning methods, aiming to develop agents that can quickly adapt and perform effectively in diverse environments.

We actively explore the intersection of RL with other fields, such as unsupervised learning and deep neural networks. By combining RL with unsupervised learning, we aim to develop methods that leverage unlabelled data and intrinsic rewards to enable agents to learn meaningful representations and extract useful knowledge from raw sensory inputs. Additionally, they explore the integration of deep neural networks within RL frameworks, enabling agents to handle high-dimensional inputs, learn complex mappings, and make informed decisions based on rich sensory information.

Our research delves into unsupervised and reinforcement learning, seeking innovative methods to leverage unlabelled data and reinforcement signals to drive learning and decision-making processes. This entails developing unsupervised learning algorithms that can autonomously discover meaningful representations and structures in data, as well as reinforcement learning techniques that enable intelligent agents to learn from interaction with the environment and optimize complex tasks.