I’m Kolby Nottingham, a computer science PhD candidate at UC Irvine, co-advised by Roy Fox and Sameer Singh. My research lies at the intersection of reinforcement learning and natural language processing. In the past, I have worked on natural language instruction following, empowering RL agents with the ability query with LLMs, using LLMs to guide exploration for RL agents, and continual in-context policy improvement. I am currently exploring using reinforcement learning to improve LLMs via conversation-level feedback signals. Stay tuned for future work in these and other exciting areas!
I’ve also previously worked with BYU’s Perception Control Cognition lab and USC’s Verification Intelligence Design Analysis group, interned with Nvidia’s Applied Deep Learning Research group and Unity’s ML-agents team, and collaborated and interned with researchers from AI2‘s mosaic and aristo teams. More details on these and other projects can be found on my experience and research pages.
News
- I’ll be interning at Riot Games this summer in Tech Research’s Generative AI team
- Our paper Selective Perception: Optimizing State Descriptions with Reinforcement Learning for Language Model Actors was accepted to NAACL 2024
- I gave a guest lecture in UCI’s graduate-level RL course on all things RLHF
- Our preprint Skill Set Optimization: Reinforcing Language Model Behavior via Transferable Skills is available on arxiv
- Our work Do Embodied Agents Dream of Pixelated Sheep?: Embodied Decision Making using Language Guided World Modelling was accepted at ICML 2023