papersSEP 12 04:00 UTC
Conformal Calibration Approach Proposed for Certified Dataset Ownership Verification
A new arXiv paper introduces CertDW, a method for verifying whether a public dataset was used to train a given deep neural network. The approach applies conformal calibration to produce certified ownership decisions, addressing weaknesses the authors identify in existing dataset ownership verification techniques. The work targets copyright protection for open-source datasets such as ImageNet.