Most of the existing works use projection functions for ternary quantization in discrete space. Scaling factors and thresholds are used in some cases to improve the model accuracy. However, the gradients used for optimization are inaccurate and result in a notable accuracy gap between the full precision and ternary models. To get more accurate gradients, some works gradually increase the discrete portion of the full precision weights in the forward propagation pass, e.g., using temperature-based Sigmoid function. Instead of directly performing ternary quantization in discrete space, we push full precision weights close to ternary ones through regularization term prior to ternary quantization. In addition, inspired by the temperature-based method, we introduce a re-scaling factor to obtain more accurate gradients by simulating the derivatives of Sigmoid function. The experimental results show that our method can significantly improve the accuracy of ternary quantization in both image classification and object detection tasks.

Model quantization enables the deployment of deep neural networks under resource-constrained devices. Vector quantization aims at reducing the model size by indexing model weights with full-precision embeddings, i.e., codewords, while the index needs to be restored to 32-bit during computation. Binary and other low-precision quantization methods can reduce the model size up to 32$\times$, however, at the cost of a considerable accuracy drop. In this paper, we propose an efficient framework for ternary quantization to produce smaller and more accurate compressed models. By integrating hyperspherical learning, pruning and reinitialization, our proposed Hyperspherical Quantization (HQ) method reduces the cosine distance between the full-precision and ternary weights, thus reducing the bias of the straight-through gradient estimator during ternary quantization. Compared with existing work at similar compression levels ($\sim$30$\times$, $\sim$40$\times$), our method significantly improves the test accuracy and reduces the model size.

Inference time, model size, and accuracy are three key factors in deep model compression. Most of the existing work addresses these three key factors separately as it is difficult to optimize them all at the same time. For example, low-bit quantization aims at obtaining a faster model; weight sharing quantization aims at improving compression ratio and accuracy; and mixed-precision quantization aims at balancing accuracy and inference time. To simultaneously optimize bit-width, model size, and accuracy, we propose pruning ternary quantization (PTQ): a simple, effective, symmetric ternary quantization method. We integrate L2 normalization, pruning, and the weight decay term to reduce the weight discrepancy in the gradient estimator during quantization, thus producing highly compressed ternary weights. Our method brings the highest test accuracy and the highest compression ratio. For example, it produces a 939kb (49$\times$) 2bit ternary ResNet-18 model with only 4\% accuracy drop on the ImageNet dataset. It compresses 170MB Mask R-CNN to 5MB (34$\times$) with only 2.8\% average precision drop. Our method is verified on image classification, object detection/segmentation tasks with different network structures such as ResNet-18, ResNet-50, and MobileNetV2.

Most existing pruning works are resource-intensive, requiring retraining or fine-tuning of the pruned models for accuracy. We propose a retraining-free pruning method based on hyperspherical learning and loss penalty terms. The proposed loss penalty term pushes some of the model weights far from zero, while the rest weight values are pushed near zero and can be safely pruned with no need for retraining and a negligible accuracy drop. In addition, our proposed method can instantly recover the accuracy of a pruned model by replacing the pruned values with their mean value. Our method obtains state-of-the-art results in retraining-free pruning and is evaluated on ResNet-18/50 and MobileNetV2 with ImageNet dataset. One can easily get a 50\% pruned ResNet18 model with a 0.47\% accuracy drop. With fine-tuning, the experiment results show that our method can significantly boost the accuracy of the pruned models compared with existing works. For example, the accuracy of a 70\% pruned (except the first convolutional layer) MobileNetV2 model only drops 3.5\%, much less than the 7\% $\sim$ 10\% accuracy drop with conventional methods.

为了以计算有效的方式部署深层模型，经常使用模型量化方法。此外，由于新的硬件支持混合的位算术操作，最近对混合精度量化（MPQ）的研究开始通过搜索网络中不同层和模块的优化位低宽，从而完全利用表示的能力。但是，先前的研究主要是在使用强化学习，神经体系结构搜索等的昂贵方案中搜索MPQ策略，或者简单地利用部分先验知识来进行位于刻度分配，这可能是有偏见和优势的。在这项工作中，我们提出了一种新颖的随机量化量化（SDQ）方法，该方法可以在更灵活，更全球优化的空间中自动学习MPQ策略，并具有更平滑的梯度近似。特别是，可区分的位宽参数（DBP）被用作相邻位意选择之间随机量化的概率因素。在获取最佳MPQ策略之后，我们将进一步训练网络使用熵感知的bin正则化和知识蒸馏。我们广泛评估了不同硬件（GPU和FPGA）和数据集的多个网络的方法。 SDQ的表现优于所有最先进的混合或单个精度量化，甚至比较低的位置量化，甚至比各种重新网络和Mobilenet家族的全精度对应物更好，这表明了我们方法的有效性和优势。

由于其不断增加的资源需求，在低资源边缘设备上部署深层神经网络是具有挑战性的。最近的研究提出了无倍数的神经网络，以减少计算和记忆消耗。 Shift神经网络是这些减少的最有效工具之一。但是，现有的低位换档网络不如其完整的精度对应物准确，并且由于其固有的设计缺陷，无法有效地转移到广泛的任务中。我们提出了利用以下新颖设计的光泽网络。首先，我们证明低位移位网络中的零重量值既不有用，也不简化模型推断。因此，我们建议使用零移动机制来简化推理，同时增加模型容量。其次，我们设计了一个新的指标，以测量训练低位移位网络中的重量冻结问题，并提出一个符号尺度分解以提高训练效率。第三，我们提出了低变化的随机初始化策略，以提高模型在转移学习方案中的性能。我们对各种计算机视觉和语音任务进行了广泛的实验。实验结果表明，光泽网络明显胜过现有的低位乘法网络，并可以实现全精度对应物的竞争性能。它还表现出强大的转移学习表现，没有准确性下降。

Mixed-precision quantization has been widely applied on deep neural networks (DNNs) as it leads to significantly better efficiency-accuracy tradeoffs compared to uniform quantization. Meanwhile, determining the exact precision of each layer remains challenging. Previous attempts on bit-level regularization and pruning-based dynamic precision adjustment during training suffer from noisy gradients and unstable convergence. In this work, we propose Continuous Sparsification Quantization (CSQ), a bit-level training method to search for mixed-precision quantization schemes with improved stability. CSQ stabilizes the bit-level mixed-precision training process with a bi-level gradual continuous sparsification on both the bit values of the quantized weights and the bit selection in determining the quantization precision of each layer. The continuous sparsification scheme enables fully-differentiable training without gradient approximation while achieving an exact quantized model in the end.A budget-aware regularization of total model size enables the dynamic growth and pruning of each layer's precision towards a mixed-precision quantization scheme of the desired size. Extensive experiments show CSQ achieves better efficiency-accuracy tradeoff than previous methods on multiple models and datasets.

由于稀疏神经网络通常包含许多零权重，因此可以在不降低网络性能的情况下潜在地消除这些不必要的网络连接。因此，设计良好的稀疏神经网络具有显着降低拖鞋和计算资源的潜力。在这项工作中，我们提出了一种新的自动修剪方法 - 稀疏连接学习（SCL）。具体地，重量被重新参数化为可培训权重变量和二进制掩模的元素方向乘法。因此，由二进制掩模完全描述网络连接，其由单位步进函数调制。理论上，从理论上证明了使用直通估计器（STE）进行网络修剪的基本原理。这一原则是STE的代理梯度应该是积极的，确保掩模变量在其最小值处收敛。在找到泄漏的Relu后，SoftPlus和Identity Stes可以满足这个原理，我们建议采用SCL的身份STE以进行离散面膜松弛。我们发现不同特征的面具梯度非常不平衡，因此，我们建议将每个特征的掩模梯度标准化以优化掩码变量训练。为了自动训练稀疏掩码，我们将网络连接总数作为我们的客观函数中的正则化术语。由于SCL不需要由网络层设计人员定义的修剪标准或超级参数，因此在更大的假设空间中探讨了网络，以实现最佳性能的优化稀疏连接。 SCL克服了现有自动修剪方法的局限性。实验结果表明，SCL可以自动学习并选择各种基线网络结构的重要网络连接。 SCL培训的深度学习模型以稀疏性，精度和减少脚波特的SOTA人类设计和自动修剪方法训练。

Embedding tables are usually huge in click-through rate (CTR) prediction models. To train and deploy the CTR models efficiently and economically, it is necessary to compress their embedding tables at the training stage. To this end, we formulate a novel quantization training paradigm to compress the embeddings from the training stage, termed low-precision training (LPT). Also, we provide theoretical analysis on its convergence. The results show that stochastic weight quantization has a faster convergence rate and a smaller convergence error than deterministic weight quantization in LPT. Further, to reduce the accuracy degradation, we propose adaptive low-precision training (ALPT) that learns the step size (i.e., the quantization resolution) through gradient descent. Experiments on two real-world datasets confirm our analysis and show that ALPT can significantly improve the prediction accuracy, especially at extremely low bit widths. For the first time in CTR models, we successfully train 8-bit embeddings without sacrificing prediction accuracy. The code of ALPT is publicly available.

训练后量化（PTQ）由于其在部署量化的神经网络方面的便利性而引起了越来越多的关注。 Founding是量化误差的主要来源，仅针对模型权重进行了优化，而激活仍然使用圆形至最终操作。在这项工作中，我们首次证明了精心选择的激活圆形方案可以提高最终准确性。为了应对激活舍入方案动态性的挑战，我们通过简单的功能适应圆形边框，以在推理阶段生成圆形方案。边界函数涵盖了重量误差，激活错误和传播误差的影响，以消除元素误差的偏差，从而进一步受益于模型的准确性。我们还使边境意识到全局错误，以更好地拟合不同的到达激活。最后，我们建议使用Aquant框架来学习边界功能。广泛的实验表明，与最先进的作品相比，Aquant可以通过可忽略不计的开销来取得明显的改进，并将Resnet-18的精度提高到2位重量和激活后训练后量化下的精度最高60.3 \％。

量化的神经网络通常需要较小的内存占用和较低的计算复杂性，这对于有效部署至关重要。然而，量化不可避免地导致原始网络的分布分发，这通常会降低性能。为了解决这个问题，已经制定了大规模的努力，但大多数现有方法缺乏统计因素，依赖于几种手动配置。在本文中，我们提出了一种自适应映射量化方法，以学习模型内固有的最佳潜在子分布，并用混凝土高斯混合物（GM）平稳地近似。特别地，网络权重被符合GM - 近似的子分布。该子分布随着直接任务客观优化引导的共同调整模式中的重量更新而发展。在各种现代架构上的图像分类和物体检测的充分实验证明了所提出的方法的有效性，泛化性能和可转移性。此外，开发了用于移动CPU的有效部署流，在Octa-Core ARM CPU上实现高达7.46 $ \ Times $推理加速。代码在https://github.com/runpeidong/dgms公开发布。

While machine learning is traditionally a resource intensive task, embedded systems, autonomous navigation, and the vision of the Internet of Things fuel the interest in resource-efficient approaches. These approaches aim for a carefully chosen trade-off between performance and resource consumption in terms of computation and energy. The development of such approaches is among the major challenges in current machine learning research and key to ensure a smooth transition of machine learning technology from a scientific environment with virtually unlimited computing resources into everyday's applications. In this article, we provide an overview of the current state of the art of machine learning techniques facilitating these real-world requirements. In particular, we focus on deep neural networks (DNNs), the predominant machine learning models of the past decade. We give a comprehensive overview of the vast literature that can be mainly split into three non-mutually exclusive categories: (i) quantized neural networks, (ii) network pruning, and (iii) structural efficiency. These techniques can be applied during training or as post-processing, and they are widely used to reduce the computational demands in terms of memory footprint, inference speed, and energy efficiency. We also briefly discuss different concepts of embedded hardware for DNNs and their compatibility with machine learning techniques as well as potential for energy and latency reduction. We substantiate our discussion with experiments on well-known benchmark datasets using compression techniques (quantization, pruning) for a set of resource-constrained embedded systems, such as CPUs, GPUs and FPGAs. The obtained results highlight the difficulty of finding good trade-offs between resource efficiency and predictive performance.

在本文中，我们介绍了一种新颖的神经网络重量压缩方法。在我们的方法中，我们将重量张量存储为稀疏，量化的矩阵因子，其产品在推理过程中即时计算以生成目标模型的权重。我们使用预计的梯度下降方法来找到重量张量的量化和稀疏分解。我们表明，这种方法可以看作是重量SVD，矢量量化和稀疏PCA的统一。结合端到端微调，我们的方法超出了或与以前的最先进方法相提并论，就精度和模型大小之间的权衡而言。我们的方法适用于中等压缩方案，与矢量量化和极端压缩方案不同。

通过移除昂贵的乘法操作并将连续权重量化成低比特离散值来减少计算复杂性，与传统的神经网络相比，这是快速且节能的低比特离散值。然而，现有的换档网络对重量初始化敏感，并且还产生由消失梯度和重量率冻结问题引起的降级性能。为了解决这些问题，我们提出了一种低点重新参数化，这是一种用于训练低位换档网络的新技术。我们的方法以符号稀疏偏移3倍的方式分解离散参数。以这种方式，它有效地学习了一个低比特网络，其权重动力学类似于全精密网络并对重量初始化不敏感。我们所提出的培训方法推动移位神经网络的界限，并以在想象中的前1个精度方面显示出3位换档网络。

In this paper, we develop a new algorithm, Annealed Skewed SGD - AskewSGD - for training deep neural networks (DNNs) with quantized weights. First, we formulate the training of quantized neural networks (QNNs) as a smoothed sequence of interval-constrained optimization problems. Then, we propose a new first-order stochastic method, AskewSGD, to solve each constrained optimization subproblem. Unlike algorithms with active sets and feasible directions, AskewSGD avoids projections or optimization under the entire feasible set and allows iterates that are infeasible. The numerical complexity of AskewSGD is comparable to existing approaches for training QNNs, such as the straight-through gradient estimator used in BinaryConnect, or other state of the art methods (ProxQuant, LUQ). We establish convergence guarantees for AskewSGD (under general assumptions for the objective function). Experimental results show that the AskewSGD algorithm performs better than or on par with state of the art methods in classical benchmarks.

当通过模拟量化训练神经网络时，我们观察到，量化的权重可以意外地在两个网格点之间振荡。这种效果的重要性及其对量化感知培训（QAT）的影响并未在文献中得到充分理解或研究。在本文中，我们更深入地研究了重量振荡现象，并表明由于推理过程中错误估计的批次纳入统计量和训练期间的噪声增加，它可能导致明显的准确性降解。这些效果在低位（$ \ leq $ 4位）的高效网络中尤其明显，具有深度可分开的层，例如mobilenets和效率网络。在我们的分析中，我们研究了一些先前提出的QAT算法，并表明其中大多数无法克服振荡。最后，我们提出了两种新型的QAT算法来克服训练期间的振荡：振荡衰减和迭代重量冻结。我们证明，我们的算法对于低位（3＆4位）的重量（3＆4位）的最新精度以及有效体系结构的激活量化，例如MobilenetV2，MobilenetV3和Imagenet上的EfficentNet-Lite。我们的源代码可在{https://github.com/qualcomm-ai-research/oscillations-qat}上获得。

深度神经网络（DNN）在解决许多真实问题方面都有效。较大的DNN模型通常表现出更好的质量（例如，精度，精度），但它们的过度计算会导致长期推理时间。模型稀疏可以降低计算和内存成本，同时保持模型质量。大多数现有的稀疏算法是单向移除的重量，而其他人则随机或贪婪地探索每层进行修剪的小权重子集。这些算法的局限性降低了可实现的稀疏性水平。此外，许多算法仍然需要预先训练的密集模型，因此遭受大的内存占地面积。在本文中，我们提出了一种新颖的预定生长和修剪（间隙）方法，而无需预先培训密集模型。它通过反复生长一个层次的层来解决以前的作品的缺点，然后在一些训练后修剪回到稀疏。实验表明，使用所提出的方法修剪模型匹配或击败高度优化的密集模型的质量，在各种任务中以80％的稀疏度，例如图像分类，客观检测，3D对象分段和翻译。它们还优于模型稀疏的其他最先进的（SOTA）方法。作为一个例子，通过间隙获得的90％不均匀的稀疏resnet-50模型在想象中实现了77.9％的前1个精度，提高了先前的SOTA结果1.5％。所有代码将公开发布。

模型二进制化是一种压缩神经网络并加速其推理过程的有效方法。但是，1位模型和32位模型之间仍然存在显着的性能差距。实证研究表明，二进制会导致前进和向后传播中的信息损失。我们提出了一个新颖的分布敏感信息保留网络（DIR-NET），该网络通过改善内部传播和引入外部表示，将信息保留在前后传播中。 DIR-NET主要取决于三个技术贡献：（1）最大化二进制（IMB）的信息：最小化信息损失和通过重量平衡和标准化同时同时使用权重/激活的二进制误差； （2）分布敏感的两阶段估计器（DTE）：通过共同考虑更新能力和准确的梯度来通过分配敏感的软近似来保留梯度的信息； （3）代表性二进制 - 意识蒸馏（RBD）：通过提炼完整精确和二元化网络之间的表示来保留表示信息。 DIR-NET从统一信息的角度研究了BNN的前进过程和后退过程，从而提供了对网络二进制机制的新见解。我们的DIR-NET中的三种技术具有多功能性和有效性，可以在各种结构中应用以改善BNN。关于图像分类和客观检测任务的综合实验表明，我们的DIR-NET始终优于主流和紧凑型体系结构（例如Resnet，vgg，vgg，EfficityNet，darts和mobilenet）下最新的二进制方法。此外，我们在现实世界中的资源有限设备上执行DIR-NET，该设备可实现11.1倍的存储空间和5.4倍的速度。

Although weight and activation quantization is an effective approach for Deep Neural Network (DNN) compression and has a lot of potentials to increase inference speed leveraging bit-operations, there is still a noticeable gap in terms of prediction accuracy between the quantized model and the full-precision model. To address this gap, we propose to jointly train a quantized, bit-operation-compatible DNN and its associated quantizers, as opposed to using fixed, handcrafted quantization schemes such as uniform or logarithmic quantization. Our method for learning the quantizers applies to both network weights and activations with arbitrary-bit precision, and our quantizers are easy to train. The comprehensive experiments on CIFAR-10 and ImageNet datasets show that our method works consistently well for various network structures such as AlexNet, VGG-Net, GoogLeNet, ResNet, and DenseNet, surpassing previous quantization methods in terms of accuracy by an appreciable margin. Code available at https://github.com/Microsoft/LQ-Nets

Quantization has become a predominant approach for model compression, enabling deployment of large models trained on GPUs onto smaller form-factor devices for inference. Quantization-aware training (QAT) optimizes model parameters with respect to the end task while simulating quantization error, leading to better performance than post-training quantization. Approximation of gradients through the non-differentiable quantization operator is typically achieved using the straight-through estimator (STE) or additive noise. However, STE-based methods suffer from instability due to biased gradients, whereas existing noise-based methods cannot reduce the resulting variance. In this work, we incorporate exponentially decaying quantization-error-aware noise together with a learnable scale of task loss gradient to approximate the effect of a quantization operator. We show this method combines gradient scale and quantization noise in a better optimized way, providing finer-grained estimation of gradients at each weight and activation layer's quantizer bin size. Our controlled noise also contains an implicit curvature term that could encourage flatter minima, which we show is indeed the case in our experiments. Experiments training ResNet architectures on the CIFAR-10, CIFAR-100 and ImageNet benchmarks show that our method obtains state-of-the-art top-1 classification accuracy for uniform (non mixed-precision) quantization, out-performing previous methods by 0.5-1.2% absolute.