chAIn--a preliminary realization of a centerless artificial intelligence
This article is extracted from https://github.com/Riksi/chAIn
What is chAIn?
The goal of the chAIn project is to develop an AI + Blockchain-enabled decentralized data and model bank.
Capitalize data and models. Initially, we hope to focus on developing a true peer-to-peer data and model lending platform to integrate the interests of different entities. Relying on ordinary smart contracts can do this, but there are still risks. Smart contracts can replace trusted intermediaries themselves, but usually these intermediaries (such as financial institutions and law firms) also provide professional knowledge, and this knowledge may not be copied by ordinary smart contracts . However, machine learning algorithms are increasingly capable of performing many professional tasks. Fintech companies such as Challenger Bank often rely on large amounts of data to provide other types of products and services, such as loans to a more diverse customer base. ML algorithms trained on the financial data of many users can be used to determine the types of products offered to new users. However, it is important that users whose data is used to build algorithm models also need some kind of compensation .
Our goal is to combine these two cutting-edge technologies, disperse in this service today , and guide the intelligentization of value where needed by third parties with as little intervention as possible . We were inspired by the OpenMined project and used its open source code base.
The key to the OpenMined project is a method called homomorphic encryption, which allows local training of machine learning algorithms on user devices without revealing the details of the algorithm so that malicious actors cannot modify it while preserving the privacy of user data. OpenMined is a good project. The encryption and privacy protection deep learning library PySyft has about 1000 stars. It uses trusted multi-party computing (Multi Party Computation) to weaken the association between private data and model training.
In addition, the use of smart solution to get a good contract data for the ML model problem - such as large public data sets financial / health information and other sensitive data are not readily available, in addition to the quality of the data may not be good. Smart contracts can be used not only to encourage people to share data, but also to encourage people to share high-quality data through rewards (that is, model accuracy improves as data quality improves ).
