ELMAENet: A Simple, Effective and Fast Deep Architecture for Image Classification

摘要

Deep learning has drawn extensive attention in machine learning because of its excellent performance, especially the convolutional neural network (CNN) architecture for image classification task. Therefore, many variant deep models based on CNN have been proposed in the past few years. However, the success of these models depends mostly on fine-tuning using backpropagation, which is a time-consuming process and suffers from troubles including slow convergence rate, local minima, intensive human intervention,etc. And these models achieve excellent performance only when their architectures are deeper enough. To overcome the above problems, we propose a simple, effective and fast deep architecture called ELMAENet, which uses extreme learning machines auto-encoder (ELM-AE) to get the filters of convolutional layer. ELMAENet incorporates the power of convolutional layer and ELM-AE (Kasun et al. in IEEE Intell Syst 28(6):31–34, 2013), which no longer need parameter tuning but still has a good performance for image classification. Experiments on several datasets have shown that the proposed ELMAENet achieves comparable or even better performance than that of the state-of-the-art models.