Research

How to Create a Mind. Our research aims to develop general-purpose autonomous agent that can solve the difficult problems of humanity. We aim to achieve this by creating learning algorithms that are fundamental to perception, cognition, and action. We hope that this computational approach will also provide insights into the inner workings of the human mind.

We are particularly interested in learning generative world models, representations, high-level cognition mechanisms, and policies that can be self-supervised, structured, and systematically generalized. This endeavor may involve exploring generative models, causality, compositionality, as well as temporal and hierarchical structures. Some examples of currently active projects are:

Unsupervised Structured Representation for High-Level Cognition Visual Reasoning & Abstraction & Systematic Generalization World Models and Planning for Model-Based Reinforcement Learning LLM-Augmented Generalist Agent in Open-Ended Environments Value Alignment and AI Safety

We tackle this problem using the tools of deep learning, reinforcement learning, and probabilistic generative learning. To draw insights from cognitive science and neuroscience, we also strive to understand the learning and memory algorithms of the brain.

We are currently interested in applying these algorithms to Agent Learning applications such as

Embodied AI Agents: (Robot Learning & Game AI) LLM Agents

Please feel free to contact us if you’re enthusiastic about joining us on this exciting journey.