AutoSR: Automatic Symbolic Regression by Searching Research States
AutoSR keeps a running research record for each candidate equation, then uses that record to guide and justify symbolic regression.
The paper argues that finite, noisy data can produce formulas that fit equally well while behaving differently outside the observed range. AutoSR pairs candidate equations with reasoning, computational evidence, and independent review inside a “Research State.” Proposer-reviewer agents search those states with progressive-widening Monte Carlo tree search, then synthesize the leading relation into a final report. The authors say it recovered algebraically equivalent relations on all nine selected benchmark challenges, including three cp3-bench problems no published system had recovered. ArXiv · AI/CL/LG's note
The paper argues that finite, noisy data can produce formulas that fit equally well while behaving differently outside the observed range. AutoSR pairs candidate equations with reasoning, computational evidence, and independent review inside a “Research State.” Proposer-reviewer agents search those states with progressive-widening Monte Carlo tree search, then synthesize the leading relation into a final report. The authors say it recovered algebraically equivalent relations on all nine selected benchmark challenges, including three cp3-bench problems no published system had recovered. ArXiv · AI/CL/LG's note
score 5