English

Models That Prove Their Own Correctness

Machine Learning 2025-12-19 v4 Computational Complexity Software Engineering

Abstract

How can we trust the correctness of a learned model on a particular input of interest? Model accuracy is typically measured on average over a distribution of inputs, giving no guarantee for any fixed input. This paper proposes a theoretically-founded solution to this problem: to train Self-Proving models that prove the correctness of their output to a verification algorithm VV via an Interactive Proof. Self-Proving models satisfy that, with high probability over an input sampled from a given distribution, the model generates a correct output and successfully proves its correctness to VV. The soundness property of VV guarantees that, for every input, no model can convince VV of the correctness of an incorrect output. Thus, a Self-Proving model proves correctness of most of its outputs, while all incorrect outputs (of any model) are detected by VV. We devise and analyze two generic methods for learning Self-Proving models: Transcript Learning (TL) which relies on access to transcripts of accepting interactions, and Reinforcement Learning from Verifier Feedback (RLVF) which trains a model by emulating interactions with the verifier.

Keywords

Cite

@article{arxiv.2405.15722,
  title  = {Models That Prove Their Own Correctness},
  author = {Noga Amit and Shafi Goldwasser and Orr Paradise and Guy Rothblum},
  journal= {arXiv preprint arXiv:2405.15722},
  year   = {2025}
}

Comments

NeurIPS 2025

R2 v1 2026-06-28T16:39:17.746Z