Abstract:This paper studies the Binary Neural Networks (BNNs) in which weights and activations are both binarized into 1-bit values, thus greatly reducing the memory usage and computational complexity. Since the modern deep neural networks are of sophisticated design with complex architecture for the accuracy reason, the diversity on distributions of weights and activations is very high. Therefore, the conventional sign function cannot be well used for effectively binarizing full-precision values in BNNs. To this end, we present a simple yet effective approach called AdaBin to adaptively obtain the optimal binary sets $\{b_1, b_2\}$ ($b_1, b_2\in \mathbb{R}$) of weights and activations for each layer instead of a fixed set (\textit{i.e.}, $\{-1, +1\}$). In this way, the proposed method can better fit different distributions and increase the representation ability of binarized features. In practice, we use the center position and distance of 1-bit values to define a new binary quantization function. For the weights, we propose an equalization method to align the symmetrical center of binary distribution to real-valued distribution, and minimize the Kullback-Leibler divergence of them. Meanwhile, we introduce a gradient-based optimization method to get these two parameters for activations, which are jointly trained in an end-to-end manner. Experimental results on benchmark models and datasets demonstrate that the proposed AdaBin is able to achieve state-of-the-art performance. For instance, we obtain a 66.4% Top-1 accuracy on the ImageNet using ResNet-18 architecture, and a 69.4 mAP on PASCAL VOC using SSD300. The PyTorch code is available at \url{https://github.com/huawei-noah/Efficient-Computing/tree/master/BinaryNetworks/AdaBin} and the MindSpore code is available at \url{https://gitee.com/mindspore/models/tree/master/research/cv/AdaBin}.

A Novel Binary Neural Network with Enhanced Dense Connection

Hyperbolic Binary Neural Network

BCNN: Binary Complex Neural Network

Network Binarization via Contrastive Learning

RBCN: Rectified Binary Convolutional Networks for Enhancing the Performance of 1-bit DCNNs

MeliusNet: an Improved Network Architecture for Binary Neural Networks

AdaBin: Improving Binary Neural Networks with Adaptive Binary Sets

A comprehensive review of Binary Neural Network

NAS-BNN: Neural Architecture Search for Binary Neural Networks

A&B BNN: Add&Bit-Operation-Only Hardware-Friendly Binary Neural Network

Increasing Information Entropy of Both Weights and Activations for the Binary Neural Networks

Sub-bit Neural Networks: Learning to Compress and Accelerate Binary Neural Networks

BCMLP: Binary-connected multilayer perceptrons

BiDense: Binarization for Dense Prediction

MeliusNet: Can Binary Neural Networks Achieve MobileNet-level Accuracy?

BinaryConnect: Training Deep Neural Networks with binary weights during propagations

Boosting Binary Neural Networks via Dynamic Thresholds Learning

Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1

Recurrent Bilinear Optimization for Binary Neural Networks

SA-BNN: State-Aware Binary Neural Network.

SecBNN: Efficient Secure Inference on Binary Neural Networks