Secure, Decentralized, Privacy Preserving Machine Learning System Implementation over Blockchain

Abstract

The traditional approach to centralized machine learning has transparency concerns. The future of machine learning is decentralized machine learning. Thus, many technological advance companies including Microsoft are also investing in researching approaches to decentralization in machine learning. With the upliftment of big data technology, designing optimized artificial intelligence algorithms is a must need. At the base of every machine learning algorithm we need data. Data is something that can be any unprocessed fact, value, text, sound or picture that is not being interpreted and analyzed. This data is not generated by just one party, multiple parties generate such data. The data will be geographically distributed amongst organizations. This pushes the need and research of distributed machine learning algorithms. In the current scenario, there is a central server which will run the machine learning algorithm and produce results, in this system obviously we need to collect all the data at that server itself. If the server is attacked then there is a problem of security of data. Also many organizations would not like to just lend their data to some third party. To address all such issues, we study all the possible ways for implementing a distributed machine learning system and propose a blockchain based distributed conservative system. Mainly, we design a Stochastic Gradient Descent (SGD) algorithm to learn a general predictive model over the trending blockchain technology, also taking care of Byzantine attack, using the within-N algorithm. Also analysis will be made on different machine learning algorithms and datasets as a part of testing, demonstrating the effectiveness of the model.