FML-Bench shows that a simple greedy hill-climber performs nearly as well as complex tree-search agents on ML research tasks, with an adaptive strategy that switches exploration modes outperforming all tested agents.
A survey on bias and fairness in machine learning.ACM computing surveys (CSUR), 54(6):1–35, 2021
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
FML-bench: A Controlled Study of AI Research Agent Strategies from the Perspective of Search Dynamics
FML-Bench shows that a simple greedy hill-climber performs nearly as well as complex tree-search agents on ML research tasks, with an adaptive strategy that switches exploration modes outperforming all tested agents.