Deep convolutional neural networks have been proven successful in multiple benchmark challenges in recent years. However, the performance improvements are heavily reliant on increasingly complex network architecture and a high number of parameters, which require ever increasing amounts of storage and memory capacity. Depthwise separable convolution (DSConv) can effectively reduce the number of required parameters through decoupling standard convolution into spatial and cross-channel convolution steps. However, the method causes a degradation of accuracy. To address this problem, we present depthwise multiception convolution, termed Multiception, which introduces layer-wise multiscale kernels to learn multiscale representations of all individual input channels simultaneously. We have carried out the experiment on four benchmark datasets, i.e. Cifar-10, Cifar-100, STL-10 and ImageNet32x32, using five popular CNN models, Multiception achieved accuracy promotion in all models and demonstrated higher accuracy performance compared to related works. Meanwhile, Multiception significantly reduces the number of parameters of standard convolution-based models by 32.48% on average while still preserving accuracy.

Over the last three years, the field of NLP has gone through a huge revolution thanks to deep learning. The leader of this revolution has been the recurrent neural network and particularly its manifestation as an LSTM. Concurrently the field of computer vision has been reshaped by convolutional neural networks. This post explores what we "text people" can learn from our friends who are doing vision. To set the stage and agree on a vocabulary, I'd like to introduce a few of the more common tasks in NLP.