ZKML, w/ Jason Morton

In this module, Nicolas Mohnblatt and Jason Morton, CEO of EZKL provide a detailed look into the design choices behind building a Zero-Knowledge Machine Learning (ZKML) system. Jason begins by explaining the core value proposition of ZKML: providing a cryptographic proof that a specific AI model was run on a given input to produce an output, with the ability to keep any of these components private. He then contrasts the needs of ZKML, which focuses on numerical computations on small, well-distributed real numbers, with general-purpose zkVMs that are optimized for cryptographic operations. The talk explores the engineering decisions involved, and covers various techniques for proving nonlinear functions, from lookup tables to non-determinism and variational arguments, as well as upcoming research on folding in this context.