Research

Our research focuses on developing machine learning methods that enable AI systems to autonomously interact with the world, learn generalizable rules from experience, and drive automatic self-evolution and scientific discovery in efficient and creative ways.

The Three A’s

Automatic AI Evolution Engine

AI for AI that improves AI.

We develop machine learning models that autonomously plan and conduct interactions with environments and data to learn from experience. Our research focuses on:

• Post-training Large Language Models (LLMs)
• Memory systems for agentic AI
• Learning from experience rather than relying solely on human data
• Self-improving AI architectures

Automatic AI Creativity Engine

Enhancing creativity in generative AI.

We work to enhance the creativity of generative machine learning models, enabling them to propose truly novel content. Our focus areas include:

• Post-training techniques for LLMs
• Fundamental mechanisms of diffusion models
• Novel content generation
• Creative problem-solving in AI systems

Automatic AI Discovery Engine for Biomedical Sciences

AI-driven scientific discovery.

We leverage self-evolving AI with improved creativity to accelerate discoveries in biomedical sciences. This includes:

• AI for drug discovery
• Genomics and epigenomics analysis
• Medical image analysis
• Clinical decision support systems

Research Themes

Our work spans fundamental machine learning research and applications in biomedical sciences:

Fundamental ML:

• Discrete neural communication
• Generative flow networks (GFlowNets)
• Multi-agent reinforcement learning
• Federated learning
• Continual learning

Biomedical Applications:

• Electronic health records analysis
• Medical imaging
• Genomic data integration
• Drug discovery and development