AI Research Agents
Reasoning, experiment selection, and efficient automated machine-learning research.
AI Research Preference Models
Our work on AI Research Preference Models compares candidate research solutions before committing to expensive evaluations. We study two approaches: an inference-only model that reasons over plans, code, and previous results, and an agentic variant that also runs small pilot experiments.
Adaptive research systems
My broader interests include test-time reasoning, learning from feedback, and continual adaptation. I aim to connect these ideas with agents that improve their research decisions through iterative experimentation and evaluation.
Selected publication
- Thomas Simon Foster et al. AI Research Preference Models. arXiv, 2026.