SMPC (Secure Multi-Party Computation) is a cryptographic protocol that enables multiple parties to jointly compute a function on their private inputs without revealing those inputs to each other — allowing collaborative ML training or inference without exposing any party's sensitive data.

SMPC for ML

• Secret Sharing: Split each value into shares distributed across parties — no single party can reconstruct the value.
• Garbled Circuits: Transform the computation into encrypted boolean circuits that parties evaluate without seeing intermediate values.
• Oblivious Transfer: One party selects a value from another party's inputs without revealing which value was selected.
• Inference: Run neural network inference on encrypted data — the model owner doesn't see the data, the data owner doesn't see the model.

Why It Matters

• Privacy: Multiple fabs can jointly train a model on their combined data without sharing proprietary process data.
• Correctness: SMPC guarantees correct computation — the result is the same as if all data were pooled.
• Overhead: SMPC is computationally expensive — 100-1000× slowdown compared to plaintext computation.

SMPC is computing on private data together — enabling collaborative ML without any party revealing their sensitive data.