Compiling to preserve our privacy
summary
A compiler turns a high‑level linear‑algebra DSL into distributed MPC assembly that runs on per‑party VMs, emitting masked primitives (uses Beaver triplets via a trusted dealer) and performing static analyses: shape inference, memory liveness, masking‑bit inference and fixed‑point transforms (uses fixed-point representation not floats). Prototype implements addition and multiplication, targets real deployments (banking credit‑scoring example), delivers ~7–8 decimal places of precision, and minimizes mask sizes to avoid blowup; a six‑person team reports a working system and open prototype code and papers for the MPC protocol.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:14The protocol
- 6:40Secure multiparty computation
- 11:00Use cases
- 14:09Direct embedding
- 16:30Components
- 19:01Masking
- 25:22Overview
- 25:34Source code
- 27:10Compiler
- 27:57Memory Management
- 31:13Our team
- 31:33The type system
- 41:09Public results
related talks
23mRecent progress in MPC-in-the-Head protocols
05mπMPC: Automatic Security Proofs for MPC Protocols
34mPrivacy Preserving Aggregation of Ad Conversions using MPC
33mInteraction Preserving Compilers for Secure Computation
15mHACCLE: Metaprogramming for Secure Multi-Party Computation
45mPrice of Active Security in Multiparty Computation
37mPETs of the world unite! Conquer the middle ground in the performance vs...
51mDistributed Zero-Knowledge Proofs and Applications to Secure Computation