abstract
Distributed Neural Training Simulation (DNTS) is a software framework designed to simulate, analyze, and visualize the training process of Deep Neural Networks (DNN) in a fully decentralized Peer-to-Peer (P2P) environment. Traditional distributed machine learning typically relies on centralized architectures, such as Parameter Servers, which inherently introduce a Single Point of Failure (SPOF) and potential network bottlenecks. DNTS overcomes these limitations by implementing an asynchronous, leaderless architecture based on the Gossip Learning protocol. Within this framework, autonomous nodes process local data partitions and iteratively exchange model weights with randomly selected peers to achieve global consensus and model convergence. Developed entirely in Scala 3 utilizing the Actor Model (via Akka Cluster) and strict Pure Functional Programming principles.
outcomes
