Definition
Plain language
An AI system for scientific computing that grows its own memory of methods as it solves problems.
As stated in the literature
An agentic scientific computing framework using a geometric factored action tree, problem fingerprinting, and similarity-weighted memory for warm-starting new problems with methods that worked on related ones.
Also called: GRAFT
Why it matters: Building up a reusable library of methods that worked on related problems is closer to how scientists actually operate than starting each task from scratch.
For example, when GRAFT-ATHENA encounters a new differential equation, it fingerprints the problem, looks up similar past problems, and warm-starts with the integration methods that worked there.
Heard on the show
“… The paper we're working from is titled "GRAFT-ATHENA: Self-Improving Agentic Teams for Autonomous Discovery and Evolutionary Numerical Algorithms," …”Episode 042 — An Agentic Scientific Computing System That Actually Remembers What It Learns