Explore a 25-minute conference talk on privacy-preserving machine learning, presented at the Association for Computing Machinery (ACM). Delve into the design and implementation of a general framework that addresses privacy concerns in data-intensive machine learning applications. Learn about the ABY3 mixed protocol framework, which offers a solution for training better models while maintaining data security. Discover key concepts such as always-encrypted data, matrix multiplication on shared data, piece-wise polynomial techniques, and linear regression on shared data. Gain insights into how this framework can be applied to neural networks and other machine learning models, balancing the need for massive data collection with privacy protection in an era where machine learning is increasingly offered as a service by major technology companies.