SWNet: Small-World Neural Networks and Rapid Convergence
This addresses the problem of high training costs for deep learning practitioners, offering a method to accelerate convergence with reduced parameters, though it appears incremental as it builds on existing network architectures.
The paper tackles the computational cost of training deep learning models by proposing SWNets, a transformation that optimizes network topology to achieve small-world connectivity, resulting in an average ~2.1x improvement in convergence speed while maintaining similar accuracy with fewer parameters.
Training large and highly accurate deep learning (DL) models is computationally costly. This cost is in great part due to the excessive number of trained parameters, which are well-known to be redundant and compressible for the execution phase. This paper proposes a novel transformation which changes the topology of the DL architecture such that it reaches an optimal cross-layer connectivity. This transformation leverages our important observation that for a set level of accuracy, convergence is fastest when network topology reaches the boundary of a Small-World Network. Small-world graphs are known to possess a specific connectivity structure that enables enhanced signal propagation among nodes. Our small-world models, called SWNets, provide several intriguing benefits: they facilitate data (gradient) flow within the network, enable feature-map reuse by adding long-range connections and accommodate various network architectures/datasets. Compared to densely connected networks (e.g., DenseNets), SWNets require a substantially fewer number of training parameters while maintaining a similar level of classification accuracy. We evaluate our networks on various DL model architectures and image classification datasets, namely, CIFAR10, CIFAR100, and ILSVRC (ImageNet). Our experiments demonstrate an average of ~2.1x improvement in convergence speed to the desired accuracy