Joint Coding and Scheduling Optimization for Distributed Learning Over Wireless Edge Networks

Unlike theoretical analysis of distributed learning (DL) in the literature, DL over wireless edge networks faces inherent dynamics/uncertainty connections and nodes, making less efficient or even inapplicable under highly dynamic networks. This article addresses these problems by leveraging recent advances coded computing deep dueling neural network architecture. By introducing structures/redundancy, a task can be completed without waiting for straggling nodes. conventional that only optimizes code structure, also requires to optimize selection/scheduling nodes with heterogeneous connections, capability, effects. However, neglecting aforementioned dynamics/uncertainty, resulting joint optimization coding scheduling minimize time turns out NP-hard. To tackle this account dynamics uncertainty we reformulate problem as Markov Decision Process design novel reinforcement algorithm employs architecture find jointly optimal scheme best set different tasks explicit information about environment nodes’ parameters. Simulations show proposed framework reduces average delay up 66% compared other approaches. The is applicable any uncertain