EditSR: Enhancing Neural Symbolic Regression via Edit-based Rectification

In three linesEditSR introduces a two-layer framework for neural symbolic regression. First layer generates expressions via autoregressive decoding, second layer (Rectifier) corrects syntactic errors through step-by-step editing. Rectifier is pretrained to maintain efficiency without restarting global search. Significant gains on complex expressions.

Summary generated by Claude — human-verified