Flag Aggregator: Scalable Distributed Training under Failures and Augmented Losses using Convex Optimization | Read Paper on Bytez