NOVA for Healthcare: Verifiable Federated Learning Datasets

Phala's TEEs ensure confidential training, but lack persistent storage. NOVA complements TEEs by providing secure vaults for datasets pre/post-TEE, with revokable group membership. This enables verifiable, multi-party workflows: upload shared data, TEE fine-tunes, store models back—auditable via on-chain metadata.

  • Uses off-chain TEE keys (Shade/Phala); no on-chain exposure.

  • getGroupKey now auto-handles token claim/checksum verify.

  • Update .env: CONTRACT_ID=nova-sdk-5.testnet, SHADE_API_URL=...

  • Run: Ensure group 'tee_demo_healthcare' exists in v2 contract.

Use Case

In healthcare (e.g., Phala's success story: https://phala.com/success-stories/healthcare-research), hospitals share encrypted records for federated learning without exposure.

In the following demo, Hospital A uploads encrypted data to NOVA, Hospital B retrieves and process the data through Phala's TEE, it uploads the output data to NOVA. Hospital A retrieves Hospital B's output data, process it, and uploads the final output to NOVA.

How NOVA Complements Private ML SDK

Phala focuses on runtime privacy ("no storage, no logs"); NOVA adds decentralized persistence: encrypt/upload via composite_upload, retrieve into TEE, output back. Focus: Data pipeline security without reinventing Phala's compute. NOVA handles pre-enclave sourcing (group auth) and post-enclave auditing (on-chain hashes/CIDs), extending Phala's quotes for full-lifecycle verifiability.

Running the Demos

  • Set .env (NEAR_PRIVATE_KEY, etc.).

  • Rust: cargo run --bin tee_federated_learning.

  • JS: ts-node demos/tee_federated_learning.ts.

Expected output:

PreviousNOVA for Governments: Confidential Document Vaults Before TEE ReviewNextNOVA for Decentralized AI: Persistent Agent State for Multi-Chain Agents

Last updated