Promoting Distributed Trust in Machine Learning and Computational Simulation via a Blockchain Network
Policy decisions in under-resourced scenarios are increasingly dependent on the outcomes of simulations and/or machine models developed remotely via distributed systems of workers. In the field of healthcare and disease modeling, the sharing, validation and verification of such models are of critical importance due to the human cost associated with sub-optimal outcomes. In this work, we present and utilize an end-to-end distributed system for simulating the spread of malaria in a fixed population, and assess the quality of results from a network of computing workers. We maintain transparent and auditable records of worker results for assessing input parameter and data quality by utilizing a blockchain system. By identifying individual faulty or anomalous workers via comparison to nearest neighbors or historical reward spaces, an optimized validation and verification process is proposed for sustaining trusted policy decisions . Our approach allows for the identification and ranking of worker consistency in order to properly allocated jobs for initial simulation and subsequent validations to ensure the promotion of trusted models.