Attesting Distributional Properties of Training Data for Machine Learning

arXiv:2308.09552v3 Announce Type: replace-cross
Abstract: The success of machine learning (ML) has been accompanied by increased concerns about its trustworthiness. Several jurisdictions are preparing ML regulatory frameworks. One such concern is ensuring that model training data has desirable distributional properties for certain sensitive attributes. For example, draft regulations indicate that model trainers are required to show that training datasets have specific distributional properties, such as reflecting diversity of the population. We propose the notion of property attestation allowing a prover …

abstract arxiv concerns cs.cr cs.lg data draft example frameworks machine machine learning model training data regulations regulatory success training training data type