Sorry, you need to enable JavaScript to visit this website.
Skip directly to content

Zero-Knowledge Machine Learning

47.00 mins
Aaron Brown
Thu 21 Sep 2023

In many ML applications either or both of the input data and model code may require shielding from public view. For example, an ML credit authorization algorithm might accept confidential data from applicants, and its internal parameters might be proprietary. How can downstream customers of the credit scores be confident that the input data were accurate and that the ML algorithm processed them correctly, without seeing either the input data or algorithm details? Or suppose a smart contract on the Ethereum blockchain requires ML processing that cannot be done on-chain both for efficiency and privacy reasons? How can an on-chain smart contract rely on an off-chain ML operation? This talk will only provide high-level answers to the mathematical questions—sketching out the approach rather than demonstrating finished tools. The main emphasis will be on how zero-knowledge machine learning can be useful in important financial applications.