Tools for Verifying Neural Models' Training Data

• https://arxiv.org/abs/2307.00682
• Dami Choi, Yonadav Shavit, David Duvenaud

In this work, we define the problem of Proof-of-Training-Data (PoTD): a protocol by which a third-party auditor (the “Verifier”) can verify which data was used to train a model.

Proof of Learning