Blockchain-Based Platform for Crowdsourcing Machine Learning Models Design and Training, while Incentivizing Continuous Improvement

It often happens that enterprises do not have enough expertise and technological resources to carry out a heavy and complex software development task. For this reason, it is common to delegate solutions design to external parties. One viable solution is Crowdsourcing, where the duty is moved to a wide audience of experts interested in working on enterprises' development problems in exchange for a monetary reward. Currently, various platforms for crowdsourcing software development are available, but they come with some important limitations: (1) a proprietary and not transparent platform needs to be run by a third trustable party; (2) the cost of publishing a task in the platform can be not affordable for many companies, making the crowdsourcing process economically inefficient; (3) once a solution has been provided by the "crowd", the task is closed and no further improved solutions are accepted. To overcome these issues and provide further features, in this paper we propose our blockchain-based platform for Crowdsourcing Machine Learning models design and training, with an incentive for the "crowd"to keep on looking for solutions capable of improving the current state of the art. After analyzing and comparing various blockchain-based platforms similar to the one hereby presented, our solution architecture is described in detail, three use case scenarios are presented, and future improvements are discussed.