DeInfoReg: A Decoupled Learning Framework for Better Training Throughput
This addresses training efficiency and gradient issues for deep learning practitioners, though it appears incremental as it builds on existing gradient flow decomposition techniques.
The paper tackles the vanishing gradient problem and low training throughput by introducing DeInfoReg, a decoupled learning framework that transforms long gradient flows into shorter ones and enables model parallelization across GPUs, achieving superior performance and better noise resistance compared to traditional backpropagation.
This paper introduces Decoupled Supervised Learning with Information Regularization (DeInfoReg), a novel approach that transforms a long gradient flow into multiple shorter ones, thereby mitigating the vanishing gradient problem. Integrating a pipeline strategy, DeInfoReg enables model parallelization across multiple GPUs, significantly improving training throughput. We compare our proposed method with standard backpropagation and other gradient flow decomposition techniques. Extensive experiments on diverse tasks and datasets demonstrate that DeInfoReg achieves superior performance and better noise resistance than traditional BP models and efficiently utilizes parallel computing resources. The code for reproducibility is available at: https://github.com/ianzih/Decoupled-Supervised-Learning-for-Information-Regularization/.