Decentralized Transfer Learning using Blockchain & IPFS for Deep Learning

The era of deep learning has enabled various dimensions of predictive and preventive healthcare in medical science. However, the applications in dimension are limited due to sparse availability of data and computing resources, also the required expertise of data understanding, modeling and training is not common yet. These limitations are forcing a barrier in the widespread use of deep learning applications in healthcare. In this research we propose a decentralized transfer learning mechanism based on smart contracts using private blockchain. The idea is to involve various stakeholders which includes data scientists, deep learning experts, dataset holders and deep learning infrastructure providers into a single trustless deep learning-based analytics system. The framework will allow sharing of data, and expertise in the field of deep learning, along with processing power to solve a single task in a decentralized manner. The sharing mechanism will be based on smart contracts to ensure the intellectual property of an individual remains protected. The framework once deployed on public blockchain can turn transfer learning into a new domain of research and business and will define a new era of shared computing for various domains. The research will directly impact people working in isolation to come together and create opportunities for better analytics and optimized predictions for scientific society.