I'm presenting three projects on self-evolving AI agents at the COEX Convention Center, Seoul from Jul 7th to 12th. All three projects revolve around investigating the ability of LLM/VLM Agents to holistically understand a problem setting (code optimization, combinatorial optimization, and theorem proving).
A benchmark for code optimization.
A benchmark evaluating the mathematical problem-solving skills of VLMs. (w/ Sabrina Reguyal)
An AlphaEvolve extension exploring an information-theoretic diversity measure to improve exploration diversity.