Model binarization is an effective method of compressing neural networks and accelerating their inference process, which enables state-of-the-art models to run on resource-limited devices. Recently, advanced binarization methods have been greatly improved by minimizing the quantization error directly in the forward process. However, a significant performance gap still exists between the 1-bit model and the 32-bit one. The empirical study shows that binarization causes a great loss of information in the forward and backward propagation which harms the performance of binary neural networks (BNNs), and the limited information representation ability of binarized parameter is one of the bottlenecks of BNN performance. We present a novel Distributionsensitive Information Retention Network (DIR-Net) to retain the information of the forward activations and backward gradients, which improves BNNs

The binary neural network, largely saving the storage and computation, serves as a promising technique for deploying deep models on resource-limited devices.However, the binarization inevitably causes severe information loss, and even worse, its discontinuity brings difficulty to the optimization of the deep network.To address these issues, a variety of algorithms have been proposed, and achieved satisfying progress in recent years. In this paper, we present a comprehensive survey of these algorithms, mainly categorized into the native solutions directly conducting binarization, and the optimized ones using techniques like minimizing the quantization error, improving the network loss function, and reducing the gradient error. We also investigate other practical aspects of binary neural networks such as the hardware-friendly design and the training tricks. Then, we give the evaluation and discussions on different tasks, including image classification, object detection and semantic segmentation. Finally, the challenges that may be faced in future research are prospected.the heavy computation and storage still inevitably limit the applications of the deep CNNs in practice. Besides, due to the huge model parameter space, the prediction of the neural networks is usually viewed as a black-box, which brings great challenges to the interpretability of CNNs. Some works like [21,22,23] empirically explore the function of each layer in the network. They visualize the feature maps extracted by different filters and view each filter as a visual unit focusing on different visual components.of the ResNet-50 [28], and meanwhile save more than 75% of parameters and 50% computational time. In the literature, approaches for compressing the deep networks can be classified into five categories: parameter pruning [26,29,30,31], parameter quantizing [32,33,34,35,36,37,38,39,40,41], low-rank parameter factorization [42,43,44,45,46], transferred/compact convolutional filters [47,48,49,50], and knowledge distillation [51,52,53,54,55,56]. The parameter pruning and quantizing mainly focus on eliminating the redundancy in the model parameters respectively by removing the redundant/uncritical ones or compressing the parameter space (e.g. , from the floating-point weights to the integer ones). Low-rank factorization applies the matrix/tensor decomposition techniques to estimate the informative parameters using the proxy ones of small size. The compact convolutional filter based approaches rely on the carefullydesigned structural convolutional filters to reduce the storage and computation complexity. The knowledge distillation methods try to distill a more compact model to reproduce the output of a larger network.Among the existing network compression techniques, quantization based one serves as a promising and fast solution that yields highly compact models compared to their floating-point counterparts, by representing the network weights with very low precision. Along this direction, the most extreme quantization is binarization, the interest...

No abstract

No abstract

The deep neural networks, such as the Deep-FSMN, have been widely studied for keyword spotting (KWS) applications. However, computational resources for these networks are significantly constrained since they usually run on-call on edge devices. In this paper, we present BiFSMN, an accurate and extreme-efficient binary neural network for KWS. We first construct a High-frequency Enhancement Distillation scheme for the binarization-aware training, which emphasizes the high-frequency information from the full-precision network's representation that is more crucial for the optimization of the binarized network. Then, to allow the instant and adaptive accuracy-efficiency trade-offs at runtime, we also propose a Thinnable Binarization Architecture to further liberate the acceleration potential of the binarized network from the topology perspective. Moreover, we implement a Fast Bitwise Computation Kernel for BiFSMN on ARMv8 devices which fully utilizes registers and increases instruction throughput to push the limit of deployment efficiency. Extensive experiments show that BiFSMN outperforms existing binarization methods by convincing margins on various datasets and is even comparable with the full-precision counterpart (e.g., less than 3% drop on Speech Commands V1-12). We highlight that benefiting from the thinnable architecture and the optimized 1-bit implementation, BiFSMN can achieve an impressive 22.3x speedup and 15.5x storage-saving on real-world edge hardware.