TrustlessNAS: Towards Trustless Network Architecture Search

Abstract

Machine learning (ML) and artificial intelligence (AI) technologies are becoming intrinsic in our every-day life. These technologies are present in our vehicles, our customer service departments, the applications in our smart-phones, our hospitals, to name a few. The tasks they are trying to solve have grown in complexity, thus, it is imperative we can trust not just the accuracy of the models, but the process used to design, train, and deploy them. This paper introduces the idea of TrustlessNAS, a platform for decentralized Network Architecture Search (NAS), which serves as the root-of-trust for the design, training, and validation process of NAS-based models. We propose an architecture that relies on permissioned blockchain technology to make the entire design process transparent, auditable, and trusted without the need to fully trust all the participants.