Projects
Goal-Conditioned Multi-Agent RL
Self-supervised goal-reaching results in multi-agent cooperation and exploration.
work
Emergent Compositional Skills in MoE VLAs
How mixture-of-experts vision-language-action models compose skills they were never explicitly trained on.
work
Probing LLM Numeracy
Modeling low-dimensional internal representations of numbers in large language models.
work
Rat Brain Phantom
Anatomically accurate brain models of small animals for verifying transcranial magnetic stimulation.
work
Hand Cursor
Control your laptop with hand gestures, à la Apple Vision Pro.
fun