Memory-Efficient Backpropagation through Large Linear Layers
This addresses memory constraints for training large neural networks, but it is incremental as it builds on existing randomized techniques.
The study tackled the memory inefficiency of backpropagation in large linear layers, such as in Transformers, by proposing a randomized matrix multiplication method that reduces memory usage with a moderate accuracy drop, demonstrated through fine-tuning RoBERTa on GLUE tasks.
In modern neural networks like Transformers, linear layers require significant memory to store activations during backward pass. This study proposes a memory reduction approach to perform backpropagation through linear layers. Since the gradients of linear layers are computed by matrix multiplications, we consider methods for randomized matrix multiplications and demonstrate that they require less memory with a moderate decrease of the test accuracy. Also, we investigate the variance of the gradient estimate induced by the randomized matrix multiplication. We compare this variance with the variance coming from gradient estimation based on the batch of samples. We demonstrate the benefits of the proposed method on the fine-tuning of the pre-trained RoBERTa model on GLUE tasks.