Sadegh Riazi on building tools for privacy-preserving machine learning and analytics based on secure multi-party computation.
Sadegh Riazi is CEO and co-founder of CipherMode Labs1, a startup building tools that enable data and machine learning teams to build and deploy models directly on encrypted data. CipherMode’s new open source project enables teams to develop and deploy machine learning algorithms using familiar tools, and thus opens up the possibility of using sensitive data in different scenarios both within an organization, and in cooperation with other organizations.
An effective data privacy and security policy will protect data in three states: at rest, in use, and in transit. While there have been other tools that allow teams to build and deploy models against encrypted data, in the past such tools have been too slow to be of practical use (even model inference was too slow). CipherMode is the first practical tool for building ML models against sensitive data.
Highlights in the video version:
- CiperCore and secure multi party computation
How advanced are the computations that are practical today?
Homomorphic encryption, models, training, and inference deployment
Secure enclaves, analytics, and privacy
Why use differential privacy?
Collaboration and federated learning
SMPC, training, and benefits of secret computation
When will TensorFlow integration be available?
What in the research world should we start paying attention to?
Reaction to the open source project CipherCore
- A video version of this conversation is available on our YouTube channel.
- CipherCore: Extracting insights from encrypted data
- Confidential Computing and Machine Learning
- Get Ready For Confidential Computing
- Ram Shankar: Securing machine learning applications
- Managing risk in machine learning
- Machine Learning Trends You Need To Know
- fastdup: Introducing a new free tool for curating image datasets at scale
 Ben Lorica is an investor in CipherMode Labs.
[Image: Blocks outside The Source in Sioux Falls by Ben Lorica.]