Definition
Plain language
A FAIR framework where AI agents arrange neural-network building blocks like Lego to find better architectures.
As stated in the literature
An agentic neural architecture search framework in which LLM agents output structured strings specifying arrangements of attention, MLP, and Mamba blocks, with the harness handling all evaluation.
Why it matters: It explores whether AI itself can find better neural-network architectures than the small set humans have settled on.
For example, an AIRA-Compose agent might propose a new model by stacking two attention blocks, one Mamba block, and an MLP in a specific order, then measure the result.
Heard on the show
“A team at FAIR — Meta's research lab — put out a paper called "Agentic Discovery of Neural Architectures: AIRA-Compose and AIRA-Design," and we're recording four days later.”Episode 053 — An AI Agent Swapped In Focal Loss And Beat A Human-Tuned Training Script