preprint · 2025-10-12
FML-bench: A Benchmark for Automatic ML Research Agents Highlighting the Importance of Exploration Breadth
On this page
Why it matters here
FML-bench evaluates iterative machine-learning research on eight fundamental tasks using existing research repositories. The authors report an association between broader exploration and better results across their tested agents. This older study adds context for research-agent strategy and evaluation, rather than evidence of a system recursively improving its own research ability.
AI-R&D capabilityEvaluation integrity
What to keep in mind
- Author-reported preprint and source extraction; no independent replication.
- Each agent has three 100-iteration rounds; the reported performance selects the best round using the test metric. Agent/backbone and evaluation protections are not uniformly matched.
- Academic-modification classification uses an automated model, rather than human validation of scientific novelty.
- This contextual record adds no benchmark scores, autonomy level or figure interpretation; original plot inspection and protocol-level numeric extraction remain pending.