Boosting Convolutional Neural Networks with Middle Spectrum Grouped Convolution

This paper proposes a novel module called middle spectrum grouped convolution (MSGC) for efficient deep convolutional neural networks (DCNNs) with the mechanism of grouped convolution. It explores the broad "middle spectrum" area between channel pruning and conventional grouped convolution. Compared with channel pruning, MSGC can retain most of the information from the input feature maps due to the group mechanism; compared with grouped convolution, MSGC benefits from the learnability, the core of channel pruning, for constructing its group topology, leading to better channel division. The middle spectrum area is unfolded along four dimensions: group-wise, layer-wise, sample-wise, and attention-wise, making it possible to reveal more powerful and interpretable structures. As a result, the proposed module acts as a booster that can reduce the computational cost of the host backbones for general image recognition with even improved predictive accuracy. For example, in the experiments on ImageNet dataset for image classification, MSGC can reduce the multiply-accumulates (MACs) of ResNet-18 and ResNet-50 by half but still increase the Top-1 accuracy by more than 1%. With 35% reduction of MACs, MSGC can also increase the Top-1 accuracy of the MobileNetV2 backbone. Results on MS COCO dataset for object detection show similar observations. Our code and trained models are available at https://github.com/hellozhuo/msgc.

A Learning Aided Gradient Descent for MISO Beamforming

This paper proposes a learning aided gradient descent (LAGD) algorithm to solve the weighted sum rate (WSR) maximization problem for multiple-input single-output (MISO) beamforming. The proposed LAGD algorithm directly optimizes the transmit precoder through implicit gradient descent based iterations, at each of which the optimization strategy is determined by a neural network, and thus, is dynamic and adaptive. At each instance of the problem, this network is initialized randomly, and updated throughout the iterative solution process. Therefore, the LAGD algorithm can be implemented at any signal-to-noise ratio (SNR) and for arbitrary antenna/user numbers, does not require labelled data or training prior to deployment. Numerical results show that the LAGD algorithm can outperform of the well-known WMMSE algorithm as well as other learning-based solutions with a modest computational complexity. Our code is available at https://github.com/XiaGroup/LAGD.