This document discusses privacy-preserving data analysis and multi-party machine learning. It introduces the challenges of jointly learning models on private data from multiple parties without sharing the raw data. Secure multi-party computation techniques allow parties to engage in online secure communications to compute functions while learning only the output and nothing else. The document presents work on a system for privacy-preserving distributed linear regression on vertically partitioned data with formal privacy guarantees. It also discusses other applications like private document classification in federated databases and privacy-preserving distributed hypothesis testing.