Protecting Sensitive Data and Machine Learning Models
summary
Shows a working pipeline that encrypts data, deploys a Python ONNX ResNet model into AWS Nitro enclaves via Cape, and runs predictions with attestation and scoped tokens; the Pi Cape SDK and experimental CLI encrypt inputs, deploy an EIF, and return top‑5 classes (example: golden retriever at 39.7%). Demonstrates key management choices, automatic client-side encryption, and encrypt-for‑another‑user flows; Nitro enclaves attestation, ONNX runtime + ResNet.
AI-written summary · about
chapters (12)
- 0:00Protecting sensitive data & ML models
- 1:14Confidential computing importance
- 2:16Encryption key challenges & solutions
- 4:00AWS Nitro Enclaves overview
- 6:12Cape introduction & components
- 7:57Pi Cape Python SDK & functions
- 9:21Demo 1: Image classification model
- 12:57Deploying model with Pi Cape CLI
- 14:25Running model on encrypted data
- 17:41Demo 2: Encrypting data for processing
- 19:54Demo 3: Encrypting data for another user
- 20:38Closing remarks & resources
related talks
40mProtecting Sensitive Data and Models for Machine Learning
23mProtecting Sensitive Data in Web Apps
47mTowards Confidential AI for the Masses!
28mGenerative AI + Proprietary Data
23mMore Secure and Confidential Computing on Linux with Nitro Enclaves
26mMachine Learning on Encrypted Data with CrypTen
09mAn introduction to Confidential Computing
36mShift-Left for LLMs: Securing the AI Model Supply Chain