Mathematical Conversations

Cryptography and Machine Learning have shared a curious history: a scientific success for one often provided an example of an impossible task for the other. Today, the goals of the two fields are aligned. Cryptographic models and tools can and should play a role in ensuring the trustworthiness of AI and machine learning and address problems such as privacy of training input, model verification and robustness of models to adversarial inputs. I will discuss three results along these lines.