Machine Learning on Encrypted Data with CrypTen
summary
A framework called CrypTen runs on PyTorch and uses secure multiparty computation to train real models on encrypted inputs and labels, with encrypted tensors stored per party and multi‑process or cloud deployment. The demo trains a logistic regressor and a small CNN on a 100‑sample MNIST split, showing training loss drop from ~2.3 to ~1.6 over 10 epochs and a held‑out accuracy of 88% accuracy for the small model. It implements encrypted tensors saved per party, uses CrypTen on PyTorch, and shows MPC training works though it is noticeably slower.
AI-written summary · about
chapters (9)
- 0:00Start
- 2:17Secure Multi-Party Computation
- 4:34Interactive Tutorial
- 8:20Secure Multi-Party Computation Works
- 10:39Tutorial for Training a Model on Encrypted Data
- 11:55Logistic Regression for Multi-Class Classification
- 12:52Training Loop
- 14:44Convolutional Neural Network
- 24:49Model Training
related talks
11mCrypTen
40mProtecting Sensitive Data and Models for Machine Learning
01h 28mPPML: Machine Learning on Data you cannot see
21mProtecting Sensitive Data and Machine Learning Models
37mOpen-source Machine Learning on Encrypted Data
52mMachine Learning over Encrypted Data with Fully Homomorphic Encryption
23mProtecting Sensitive Data in Web Apps
16mPrivacy Preserving AI