Particle Swarm optimisation for Evolving Deep Neural Networks for Image Classification by Evolving and Stacking Transferable Blocks
Bin Wang, Bing Xue, Mengjie Zhang
Abstract
Deep Convolutional Neural Networks (CNNs) have been widely used in image classification tasks, but the process of designing CNN architectures is very complex, so Neural Architecture Search (NAS), automatically searching for optimal CNN architectures, has attracted more and more research interests. However, the computational cost of NAS is often too high to be applied to real-life applications. In this paper, an efficient particle swarm optimisation method named EPSOCNN is proposed to evolve CNN architectures inspired by the idea of transfer learning. EPSOCNN successfully reduces the computation cost by minimising the search space to a single block and utilising a small subset of the training set to evaluate CNNs during the evolutionary process. Meanwhile, EPSOCNN also keeps very competitive classification accuracy by stacking the evolved block multiple times to fit the whole training dataset. The proposed EPSOCNN algorithm is evaluated on CIFAR-10 dataset and compared with 13 peer competitors including deep CNNs crafted by hand, learned by reinforcement learning methods and evolved by evolutionary computation approaches. It shows very promising results with regard to the classification accuracy, the number of parameters and the computational cost. Besides, the evolved transferable block from CIFAR-10 is transferred and evaluated on two other datasets - CIFAR-100 and SVHN. It shows promising results on both of the datasets, which demonstrate the transferability of the evolved block. All of the experiments have been performed multiple times and Student's t-test is used to compare the proposed method with peer competitors from the statistical point of view.