Dynamic Gradient Sparse Update for Edge Training
This addresses memory constraints for edge devices like microcontrollers, enabling efficient personalized model fine-tuning with data privacy, though it is incremental as it builds on existing sparse update techniques.
The paper tackles the problem of high memory usage during training on edge devices by proposing dynamic gradient sparse updates, which update only important channels and layers to reduce memory. The method achieves 85.77% accuracy on CIFAR-10 with MobileNetV2 while using only 2% of convolution weights and 256KB memory, reducing feature memory usage by 98% compared to dense training.
Training on edge devices enables personalized model fine-tuning to enhance real-world performance and maintain data privacy. However, the gradient computation for backpropagation in the training requires significant memory buffers to store intermediate features and compute losses. This is unacceptable for memory-constrained edge devices such as microcontrollers. To tackle this issue, we propose a training acceleration method using dynamic gradient sparse updates. This method updates the important channels and layers only and skips gradient computation for the less important channels and layers to reduce memory usage for each update iteration. In addition, the channel selection is dynamic for different iterations to traverse most of the parameters in the update layers along the time dimension for better performance. The experimental result shows that the proposed method enables an ImageNet pre-trained MobileNetV2 trained on CIFAR-10 to achieve an accuracy of 85.77\% while updating only 2\% of convolution weights within 256KB on-chip memory. This results in a remarkable 98\% reduction in feature memory usage compared to dense model training.