SHINE: SHaring the INverse Estimate from the forward pass for bi-level optimization and implicit models
This addresses the slow training speed of implicit models for researchers and practitioners in deep learning, offering an incremental improvement over existing methods.
The paper tackles the computational bottleneck in training implicit deep learning models like Deep Equilibrium Models (DEQs), which involve inverting large Jacobian matrices, by proposing SHINE, a method that uses quasi-Newton matrices from the forward pass to approximate the inverse Jacobian for gradient computation. The result shows significant computational cost reduction in backward passes, with SHINE outperforming in hyperparameter optimization and achieving similar performance to Jacobian-Free methods on larger-scale problems like MDEQs, albeit with a limited performance drop.
In recent years, implicit deep learning has emerged as a method to increase the effective depth of deep neural networks. While their training is memory-efficient, they are still significantly slower to train than their explicit counterparts. In Deep Equilibrium Models (DEQs), the training is performed as a bi-level problem, and its computational complexity is partially driven by the iterative inversion of a huge Jacobian matrix. In this paper, we propose a novel strategy to tackle this computational bottleneck from which many bi-level problems suffer. The main idea is to use the quasi-Newton matrices from the forward pass to efficiently approximate the inverse Jacobian matrix in the direction needed for the gradient computation. We provide a theorem that motivates using our method with the original forward algorithms. In addition, by modifying these forward algorithms, we further provide theoretical guarantees that our method asymptotically estimates the true implicit gradient. We empirically study this approach and the recent Jacobian-Free method in different settings, ranging from hyperparameter optimization to large Multiscale DEQs (MDEQs) applied to CIFAR and ImageNet. Both methods reduce significantly the computational cost of the backward pass. While SHINE has a clear advantage on hyperparameter optimization problems, both methods attain similar computational performances for larger scale problems such as MDEQs at the cost of a limited performance drop compared to the original models.