Unity is Power: Semi-Asynchronous Collaborative Training of Large-Scale Models With Structured Pruning in Resource-Limited Clients

In this work, we study to release the potential of massive heterogeneous weak computing power to collaboratively train large-scale models on dispersed datasets. In order to improve both efficiency and accuracy in resource-adaptive collaborative learning, we take the first step to consider the unstructured pruning, varying submodel architectures, knowledge loss, and straggler challenges simultaneously. We propose a novel semiasynchronous collaborative training framework, namely Co-S<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>P, with data distribution-aware structured pruning and cross-block knowledge transfer mechanism to address the above concerns. Furthermore, we provide theoretical proof that Co-S<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>P can achieve asymptotic optimal convergence rate of O(1/√ N∗EQ). Finally, we conduct extensive experiments on two types of tasks with a real-world hardware testbed including diverse IoT devices. The experimental results demonstrate that Co-S<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>P improves accuracy by up to 8.8% and resource utilization by up to 1.2× compared to state-of-the-art methods, while reducing memory consumption by approximately 22% and training time by about 24% on all resource-limited devices.