Dawn Song presents on privacy-preserving technology at SVIL 2018

UC Berkeley professor Dawn Song visited the Sutardja Center’s Silicon Valley Innovation Leadership program to introduce her startup, Oasis Labs, which touts privacy-preserving technology using blockchain.

Song said that with machine learning, banks could use more data to train better models for fraud detection. However, banks cannot currently share customer data due to privacy concerns, misaligned incentives, and other obstacles.

With a blockchain platform, Song said that it was possible to build a fraud detection smart contract that can run the machine learning algorithm while being completely encrypted.

“As we compute, we need to make sure that the sensitive data is being protected during the computation process,” Song said during her presentation. “We need to be able to build the platform that has the scalability sufficient to support this kind of workload.”

Oasis Labs is currently working on a project called Ekiden, which is working to keep costly smart contracts performing well at more affordable prices while maintaining privacy.

Song also discussed Keystone, a research project focused on building trusted execution environments (TEE) with secure hardware enclaves. The secure hardware is a mechanism to essentially compute a function as a black box — that is, outside operating systems are not able to see what’s on the inside. Secure hardware provides protection for confidentiality and integrity for the smart contract running inside.

One of Keystone’s main goals is to prevent memorization using a concept called differential privacy, which ensures that the query outcome is the same regardless of a user’s input. An attacker wouldn’t be able to tell what datapoint was used in the initial calculation, rendering the user immune to re-identification attacks.

“Secure enclave is not only useful for a blockchain setting — in fact, it’s a very important cornerstone security primitive,” Song said. “This is a platform for building new security applications that couldn’t be built otherwise for the same practical application.”


Originally published at UC Berkeley Sutardja Center.