How to Verify Consistency of Probabilistic Claims
A polynomial-time verifier can check whether a predictor’s many conditional-probability answers fit a single underlying distribution.
The paper builds an interactive PCP protocol for a probability circuit and a confidence circuit that together imply exponentially many claims. The verifier samples only a few circuit values and proof-oracle locations while interacting with one untrusted prover. The authors also show NP certificates for approximate consistency of explicit probabilistic claims, extending earlier work by Nilsson. They frame the result as a step toward predictive models that can prove their own consistency. ArXiv · AI/CL/LG's note
The paper builds an interactive PCP protocol for a probability circuit and a confidence circuit that together imply exponentially many claims. The verifier samples only a few circuit values and proof-oracle locations while interacting with one untrusted prover. The authors also show NP certificates for approximate consistency of explicit probabilistic claims, extending earlier work by Nilsson. They frame the result as a step toward predictive models that can prove their own consistency. ArXiv · AI/CL/LG's note
score 5