Efficient Deep Learning on Multi-Source Private Data
This addresses privacy concerns in sensitive applications like healthcare and finance, but appears incremental as it builds on existing privacy-preservation techniques.
The paper tackled the problem of training machine learning models on sensitive data from multiple sources without compromising privacy, and introduced Myelin, a framework that combines privacy-preservation primitives to establish a baseline performance for fully private machine learning.
Machine learning models benefit from large and diverse datasets. Using such datasets, however, often requires trusting a centralized data aggregator. For sensitive applications like healthcare and finance this is undesirable as it could compromise patient privacy or divulge trade secrets. Recent advances in secure and privacy-preserving computation, including trusted hardware enclaves and differential privacy, offer a way for mutually distrusting parties to efficiently train a machine learning model without revealing the training data. In this work, we introduce Myelin, a deep learning framework which combines these privacy-preservation primitives, and use it to establish a baseline level of performance for fully private machine learning.