Continual Learning with Scaled Gradient Projection

Abstract

In neural networks, continual learning results in gradient interference among sequential tasks, leading to catastrophic forgetting of old tasks while learning new ones. This issue is addressed in recent methods by storing the important gradient spaces for old tasks and updating the model orthogonally during new tasks. However, such restrictive orthogonal gradient updates hamper the learning capability of the new tasks resulting in sub-optimal performance. To improve new learning while minimizing forgetting, in this paper we propose a Scaled Gradient Projection (SGP) method, where we combine the orthogonal gradient projections with scaled gradient steps along the important gradient spaces for the past tasks. The degree of gradient scaling along these spaces depends on the importance of the bases spanning them. We propose an efficient method for computing and accumulating importance of these bases using the singular value decomposition of the input representations for each task. We conduct extensive experiments ranging from continual image classification to reinforcement learning tasks and report better performance with less training overhead than the state-of-the-art approaches.

Ada-QPacknet – Multi-Task Forget-Free Continual Learning with Quantization Driven Adaptive Pruning

“…For Ada-QPacknet, the learning rate is set resorting to a dynamic scheduler starting from 0.01 and decreasing over time with a minimum of 0.0001 for all datasets except for TinyImagenet (starting from 0.001) and Ima-genet100 (constant learning rate of 0.0001). The adopted models are a two-layer neural network with fully connected layers (p-MNIST), reduced AlexNet (s-CIFAR100) [40], Resnet-18 (5 datasets and Imagenet), TinyNet (TinyImagenet) in accordance with model backbones used in the WSN paper [22]. For weight initialization, we adopt the Xavier initializer.…”