AutoSR is a fully automated system for symbolic regression that searches persistent scientific investigations rather than isolated equations. It preserves a Research State coupling each candidate equation with reasoning, computational evidence, and independent review. Proposer-reviewer agents develop states under progressive-widening Monte Carlo tree search (PW-MCTS). The system was introduced in an arXiv paper published on 2026-08-17.
AutoSR (Automatic Symbolic Regression) is introduced as a fully automated system that instantiates Research-Space Symbolic Regression by searching persistent scientific investigations rather than isolated equations. The system addresses the problem that finite, noisy data often yield numerically competitive expressions with very different behavior outside the observed regime, making numerical fit and syntactic complexity insufficient for scientific credibility. AutoSR preserves a Research State that couples each candidate equation with the reasoning, computational evidence, and independent review developed along its branch. Proposer-reviewer agents develop these states under progressive-widening Monte Carlo tree search (PW-MCTS), which allocates computation across competing investigations. The accumulated research record is synthesized into a final report explaining the leading relation and the basis for its selection. The paper reports results across nine selected benchmarks.
AutoSR shifts symbolic regression from equation-level search to research-state search, retaining scientific context such as motivations and probes. The use of PW-MCTS allows dynamic allocation of compute across competing investigation branches, potentially improving exploration-exploitation trade-offs. The proposer-reviewer agent architecture introduces independent review as part of the search process, which may filter spurious expressions that fit data but lack scientific plausibility.
This work signals a trend toward AI systems that produce not just answers but auditable scientific reasoning trails. For industries relying on data-driven model discovery (e.g., materials, drug discovery, engineering), such systems could increase trust in automatically derived equations. The emphasis on independent review within the search process may influence how AI-generated scientific claims are validated in practice.
AutoSR could reduce the cost and time of discovering interpretable mathematical models from data, especially in scientific and industrial R&D. The generated research reports may serve as documentation for regulatory or peer review. However, the paper is a research preprint; commercial value depends on demonstrated performance on practical problems and integration into existing workflows.
Observable next signals include: (1) release of code or benchmark results beyond the nine selected problems; (2) application of AutoSR to real-world scientific datasets; (3) comparisons with existing symbolic regression methods on out-of-distribution generalization; (4) adoption of research-state search ideas in other automated science domains.