Large-Scale Distributed Second-Order Optimization Using Kronecker-Factored Approximate Curvature for Deep Convolutional Neural Networks
This addresses the problem of efficient large-scale training for deep learning practitioners, offering a novel optimization approach that is incremental but provides specific gains in convergence speed and batch size handling.
The paper tackles the generalization gap in large-scale distributed training of deep neural networks by proposing a second-order optimization method that matches the generalization of first-order methods while converging faster and handling larger mini-batch sizes. They achieved 75% Top-1 validation accuracy on ImageNet with ResNet-50 in 35 epochs for mini-batches under 16,384 and even with a mini-batch size of 131,072 in only 978 iterations.
Large-scale distributed training of deep neural networks suffer from the generalization gap caused by the increase in the effective mini-batch size. Previous approaches try to solve this problem by varying the learning rate and batch size over epochs and layers, or some ad hoc modification of the batch normalization. We propose an alternative approach using a second-order optimization method that shows similar generalization capability to first-order methods, but converges faster and can handle larger mini-batches. To test our method on a benchmark where highly optimized first-order methods are available as references, we train ResNet-50 on ImageNet. We converged to 75% Top-1 validation accuracy in 35 epochs for mini-batch sizes under 16,384, and achieved 75% even with a mini-batch size of 131,072, which took only 978 iterations.