Delta
Out-of-the-Box Blockchain-powered Verifiable PPC Framework
Overview
Delta is a decentralized network to perform statistical and machine learning computations jointly on data possessed by a group of data holders. The data never leaves its original position so that nobody else other than the original holder could see it.
Delta is useful in a lot of cases such as risk assessment model training using private data from several banks, or health condition prediction for patients using data from several hospitals.
By encapsulating modern Privacy-Preserving Computation(PPC) techniques inside, Delta doesn't require users to have any prior knowledge about PPC. Developers just write the computation tasks using PyTorch and Pandas, send it to the Delta network. And the rest parts are taken care of by Delta.
Delta network determines the task type by checking the distribution status of the data required by the task, and transforms the task into horizontal/vertical federated learning, or federated analytics task and executes it on the network.
The original data is never accessible to the developers. Delta integrates Blockchain and Zero Knowledge Proof so that the developer could verify that the computation is actually performed as designed on the required data.
Getting Started
Delta network is composed of several components, and all of them could be easily deployed with Docker. Before choosing the required components and starting the network, take a look at the System Overview and get a basic understanding about the network components.
To get your hands dirty and start to play with the network, check the Getting Started guide for a quick deployment of the network.
Join the Community
If you have any questions using the source code, submit issues on Github.
To get connected to other developers or want to know more about privacy-preserving computation techniques, join Delta's Slack:
