Cover Image for Latent Agent Swarms for Evolutionary Discovery and Optimization (AIDDA Seminar)
Cover Image for Latent Agent Swarms for Evolutionary Discovery and Optimization (AIDDA Seminar)
11 Going

Latent Agent Swarms for Evolutionary Discovery and Optimization (AIDDA Seminar)

Hosted by AIDDA Institute
Zoom
Registration
Welcome! To join the event, please register below.
About Event

​Join us for our next AIDDA seminar series talk!

​We have Qile Jiang, a PhD candidate in Applied Mathematics at Brown University, advised by George Em Karniadakis. His research focuses on scientific machine learning and AI for science, including collaborative multi-agent LLM systems for automated scientific discovery, LLM-guided evolutionary search, and neural operators for physical systems.

​He will be talking about his recent work, SwarmEvolve: Latent Agent Swarms for Evolutionary Discovery and Optimization:

​https://arxiv.org/abs/2611.08483

​Abstract for talk:

​LLM-guided evolutionary search uses AI agents to iteratively refine programs for mathematical and computational problems. What if the agents could share ideas without writing them out?

​In SwarmEvolve, we build hundreds of low-cost agents that work in groups, reason in latent space, and share information by passing KV caches. Each swarm studies the problem from a different angle. A novelty filter keeps the most distinct ideas by directly measuring distances between hidden states.

​On 21 AlphaEvolve-style problems in mathematics and scientific computing, SwarmEvolve running on Qwen3 8B achieves strong scores within minutes. Latent agent swarms may offer a fast and thorough way to explore the design space of scientific programs, at the scale open problems require and at a cost researchers can afford.

11 Going