Tensor decomposition methods are widely used for model compression and fast inference in convolutional neural networks (CNNs). Although many decompositions are conceivable, only CP decomposition and a few others have been applied in practice, and no extensive comparisons have been made between available methods. Previous studies have not determined how many decompositions are available, nor which of them is optimal. In this study, we first characterize a decomposition class specific to CNNs by adopting a flexible graphical notation. The class includes such well-known CNN modules as depthwise separable convolution layers and bottleneck layers, but also previously unknown modules with nonlinear activations. We also experimentally compare the tradeoff between prediction accuracy and time/space complexity for modules found by enumerating all possible decompositions, or by using a neural architecture search. We find some nonlinear decompositions outperform existing ones.
A surge in artificial intelligence and autonomous technologies have increased the demand toward enhanced edge-processing capabilities. Computational complexity and size of state-of-the-art Deep Neural Networks (DNNs) are rising exponentially with diverse network models and larger datasets. This growth limits the performance scaling and energy-efficiency of both distributed and embedded inference platforms. Embedded designs at the edge are constrained by energy and speed limitations of available processor substrates and processor to memory communication required to fetch the model coefficients. While many hardware accelerator and network deployment frameworks have been in development, a framework is needed to allow the variety of existing architectures, and those in development, to be expressed in critical parts of the flow that perform various optimization steps. Moreover, premature architecture-blind network selection and optimization diminish the effectiveness of schedule optimizations and hardware-specific mappings. In this paper, we address these issues by creating a cross-layer software-hardware design framework that encompasses network training and model compression that is aware of and tuned to the underlying hardware architecture. This approach leverages the available degrees of DNN structure and sparsity to create a converged network that can be partitioned and efficiently scheduled on the target hardware platform, minimizing data movement, and improving the overall throughput and energy. To further streamline the design, we leverage the high-level, flexible SoC generator platform based on RISC-V ROCC framework. This integration allows seamless extensions of the RISC-V instruction set and Chisel-based rapid generator design. Utilizing this approach, we implemented a silicon prototype in a 16 nm TSMC process node achieving record processing efficiency of up to 18 TOPS/W.
Model compression has emerged as an important area of research for deploying deep learning models on Internet-of-Things (IoT). However, for extremely memory-constrained scenarios, even the compressed models cannot fit within the memory of a single device and, as a result, must be distributed across multiple devices. This leads to a distributed inference paradigm in which memory and communication costs represent a major bottleneck. Yet, existing model compression techniques are not communication-aware. Therefore, we propose Network of Neural Networks (NoNN), a new distributed IoT learning paradigm that compresses a large pretrained 'teacher' deep network into several disjoint and highly-compressed 'student' modules, without loss of accuracy. Moreover, we propose a network science-based knowledge partitioning algorithm for the teacher model, and then train individual students on the resulting disjoint partitions. Extensive experimentation on five image classification datasets, for user-defined memory/performance budgets, show that NoNN achieves higher accuracy than several baselines and similar accuracy as the teacher model, while using minimal communication among students. Finally, as a case study, we deploy the proposed model for CIFAR-10 dataset on edge devices and demonstrate significant improvements in memory footprint (up to 24x), performance (up to 12x), and energy per node (up to 14x) compared to the large teacher model. We further show that for distributed inference on multiple edge devices, our proposed NoNN model results in up to 33x reduction in total latency w.r.t. a state-of-the-art model compression baseline.
Correlation filters (CF) have received considerable attention in visual tracking because of their computational efficiency. Leveraging deep features via off-the-shelf CNN models (e.g., VGG), CF trackers achieve state-of-the-art performance while consuming a large number of computing resources. This limits deep CF trackers to be deployed to many mobile platforms on which only a single-core CPU is available. In this paper, we propose to jointly compress and transfer off-the-shelf CNN models within a knowledge distillation framework. We formulate a CNN model pretrained from the image classification task as a teacher network, and distill this teacher network into a lightweight student network as the feature extractor to speed up CF trackers. In the distillation process, we propose a fidelity loss to enable the student network to maintain the representation capability of the teacher network. Meanwhile, we design a tracking loss to adapt the objective of the student network from object recognition to visual tracking. The distillation process is performed offline on multiple layers and adaptively updates the student network using a background-aware online learning scheme. Extensive experiments on five challenging datasets demonstrate that the lightweight student network accelerates the speed of state-of-the-art deep CF trackers to real-time on a single-core CPU while maintaining almost the same tracking accuracy.
Model compression has become necessary when applying neural networks (NN) into many real application tasks that can accept slightly-reduced model accuracy with strict tolerance to model complexity. Recently, Knowledge Distillation, which distills the knowledge from well-trained and highly complex teacher model into a compact student model, has been widely used for model compression. However, under the strict requirement on the resource cost, it is quite challenging to achieve comparable performance with the teacher model, essentially due to the drastically-reduced expressiveness ability of the compact student model. Inspired by the nature of the expressiveness ability in Neural Networks, we propose to use multi-segment activation, which can significantly improve the expressiveness ability with very little cost, in the compact student model. Specifically, we propose a highly efficient multi-segment activation, called Light Multi-segment Activation (LMA), which can rapidly produce multiple linear regions with very few parameters by leveraging the statistical information. With using LMA, the compact student model is capable of achieving much better performance effectively and efficiently, than the ReLU-equipped one with same model scale. Furthermore, the proposed method is compatible with other model compression techniques, such as quantization, which means they can be used jointly for better compression performance. Experiments on state-of-the-art NN architectures over the real-world tasks demonstrate the effectiveness and extensibility of the LMA.
The recent WSNet [1] is a new model compression method through sampling filterweights from a compact set and has demonstrated to be effective for 1D convolutionneural networks (CNNs). However, the weights sampling strategy of WSNet ishandcrafted and fixed which may severely limit the expression ability of the resultedCNNs and weaken its compression ability. In this work, we present a novel auto-sampling method that is applicable to both 1D and 2D CNNs with significantperformance improvement over WSNet. Specifically, our proposed auto-samplingmethod learns the sampling rules end-to-end instead of being independent of thenetwork architecture design. With such differentiable weight sampling rule learning,the sampling stride and channel selection from the compact set are optimized toachieve better trade-off between model compression rate and performance. Wedemonstrate that at the same compression ratio, our method outperforms WSNetby6.5% on 1D convolution. Moreover, on ImageNet, our method outperformsMobileNetV2 full model by1.47%in classification accuracy with25%FLOPsreduction. With the same backbone architecture as baseline models, our methodeven outperforms some neural architecture search (NAS) based methods such asAMC [2] and MNasNet [3].
Previous work showed empirically that large neural networks can be significantly reduced in size while preserving their accuracy. Model compression became a central research topic, as it is crucial for deployment of neural networks on devices with limited computational and memory resources. The majority of the compression methods are based on heuristics and offer no worst-case guarantees on the trade-off between the compression rate and the approximation error for an arbitrarily new sample. We propose the first efficient, data-independent neural pruning algorithm with a provable trade-off between its compression rate and the approximation error for any future test sample. Our method is based on the coreset framework, which finds a small weighted subset of points that provably approximates the original inputs. Specifically, we approximate the output of a layer of neurons by a coreset of neurons in the previous layer and discard the rest. We apply this framework in a layer-by-layer fashion from the top to the bottom. Unlike previous works, our coreset is data independent, meaning that it provably guarantees the accuracy of the function for any input $x\in \mathbb{R}^d$, including an adversarial one. We demonstrate the effectiveness of our method on popular network architectures. In particular, our coresets yield 90\% compression of the LeNet-300-100 architecture on MNIST while improving the accuracy.
Structured weight pruning is a representative model compression technique of DNNs to reduce the storage and computation requirements and accelerate inference. An automatic hyperparameter determination process is necessary due to the large number of flexible hyperparameters. This work proposes AutoCompress, an automatic structured pruning framework with the following key performance improvements: (i) effectively incorporate the combination of structured pruning schemes in the automatic process; (ii) adopt the state-of-art ADMM-based structured weight pruning as the core algorithm, and propose an innovative additional purification step for further weight reduction without accuracy loss; and (iii) develop effective heuristic search method enhanced by experience-based guided search, replacing the prior deep reinforcement learning technique which has underlying incompatibility with the target pruning problem. Extensive experiments on CIFAR-10 and ImageNet datasets demonstrate that AutoCompress is the key to achieve ultra-high pruning rates on the number of weights and FLOPs that cannot be achieved before. As an example, AutoCompress outperforms the prior work on automatic model compression by up to 33x in pruning rate (120x reduction in the actual parameter count) under the same accuracy. Significant inference speedup has been observed from the AutoCompress framework on actual measurements on smartphone. We release all models of this work at anonymous link: http://bit.ly/2VZ63dS.
End-to-end automatic speech recognition (ASR) models are increasingly large and complex to achieve the best possible accuracy. In this paper, we build an AutoML system that uses reinforcement learning (RL) to optimize the per-layer compression ratios when applied to a state-of-the-art attention based end-to-end ASR model composed of several LSTM layers. We use singular value decomposition (SVD) low-rank matrix factorization as the compression method. For our RL-based AutoML system, we focus on practical considerations such as the choice of the reward/punishment functions, the formation of an effective search space, and the creation of a representative but small data set for quick evaluation between search steps. Finally, we present accuracy results on LibriSpeech of the model compressed by our AutoML system, and we compare it to manually-compressed models. Our results show that in the absence of retraining our RL-based search is an effective and practical method to compress a production-grade ASR system. When retraining is possible, we show that our AutoML system can select better highly-compressed seed models compared to manually hand-crafted rank selection, thus allowing for more compression than previously possible.
Large deep neural network (DNN) models pose the key challenge to energy efficiency due to the significantly higher energy consumption of off-chip DRAM accesses than arithmetic or SRAM operations. It motivates the intensive research on model compression with two main approaches. Weight pruning leverages the redundancy in the number of weights and can be performed in a non-structured, which has higher flexibility and pruning rate but incurs index accesses due to irregular weights, or structured manner, which preserves the full matrix structure with lower pruning rate. Weight quantization leverages the redundancy in the number of bits in weights. Compared to pruning, quantization is much more hardware-friendly, and has become a "must-do" step for FPGA and ASIC implementations. This paper provides a definitive answer to the question for the first time. First, we build ADMM-NN-S by extending and enhancing ADMM-NN, a recently proposed joint weight pruning and quantization framework. Second, we develop a methodology for fair and fundamental comparison of non-structured and structured pruning in terms of both storage and computation efficiency. Our results show that ADMM-NN-S consistently outperforms the prior art: (i) it achieves 348x, 36x, and 8x overall weight pruning on LeNet-5, AlexNet, and ResNet-50, respectively, with (almost) zero accuracy loss; (ii) we demonstrate the first fully binarized (for all layers) DNNs can be lossless in accuracy in many cases. These results provide a strong baseline and credibility of our study. Based on the proposed comparison framework, with the same accuracy and quantization, the results show that non-structrued pruning is not competitive in terms of both storage and computation efficiency. Thus, we conclude that non-structured pruning is considered harmful. We urge the community not to continue the DNN inference acceleration for non-structured sparsity.
With the development of deep neural networks, the size of network models becomes larger and larger. Model compression has become an urgent need for deploying these network models to mobile or embedded devices. Model quantization is a representative model compression technique. Although a lot of quantization methods have been proposed, many of them suffer from a high quantization error caused by a long-tail distribution of network weights. In this paper, we propose a novel quantization method, called weight normalization based quantization (WNQ), for model compression. WNQ adopts weight normalization to avoid the long-tail distribution of network weights and subsequently reduces the quantization error. Experiments on CIFAR-100 and ImageNet show that WNQ can outperform other baselines to achieve state-of-the-art performance.
Neural network compression empowers the effective yet unwieldy deep convolutional neural networks (CNN) to be deployed in resource-constrained scenarios. Most state-of-the-art approaches prune the model in filter-level according to the "importance" of filters. Despite their success, we notice they suffer from at least two of the following problems: 1) The redundancy among filters is not considered because the importance is evaluated independently. 2) Cross-layer filter comparison is unachievable since the importance is defined locally within each layer. Consequently, we must manually specify layer-wise pruning ratios. 3) They are prone to generate sub-optimal solutions because they neglect the inequality between reducing parameters and reducing computational cost. Reducing the same number of parameters in different positions in the network may reduce different computational cost. To address the above problems, we develop a novel algorithm named as COP (correlation-based pruning), which can detect the redundant filters efficiently. We enable the cross-layer filter comparison through global normalization. We add parameter-quantity and computational-cost regularization terms to the importance, which enables the users to customize the compression according to their preference (smaller or faster). Extensive experiments have shown COP outperforms the others significantly. The code is released at https://github.com/ZJULearning/COP.
With the rapid scaling up of deep neural networks (DNNs), extensive research studies on network model compression such as weight pruning have been performed for improving deployment efficiency. This work aims to advance the compression beyond the weights to neuron activations. We propose the joint regularization technique which simultaneously regulates the distribution of weights and activations. By distinguishing and leveraging the significance difference among neuron responses and connections during learning, the jointly pruned network, namely \textit{JPnet}, optimizes the sparsity of activations and weights for improving execution efficiency. The derived deep sparsification of JPnet reveals more optimization space for the existing DNN accelerators dedicated for sparse matrix operations. We thoroughly evaluate the effectiveness of joint regularization through various network models with different activation functions and on different datasets. With $0.4\%$ degradation constraint on inference accuracy, a JPnet can save $72.3\% \sim 98.8\%$ of computation cost compared to the original dense models, with up to $5.2\times$ and $12.3\times$ reductions in activation and weight numbers, respectively.
Large capacity machine learning (ML) models are prone to membership inference attacks (MIAs), which aim to infer whether the target sample is a member of the target model's training dataset. The serious privacy concerns due to the membership inference have motivated multiple defenses against MIAs, e.g., differential privacy and adversarial regularization. Unfortunately, these defenses produce ML models with unacceptably low classification performances. Our work proposes a new defense, called distillation for membership privacy (DMP), against MIAs that preserves the utility of the resulting models significantly better than prior defenses. DMP leverages knowledge distillation to train ML models with membership privacy. We provide a novel criterion to tune the data used for knowledge transfer in order to amplify the membership privacy of DMP. Our extensive evaluation shows that DMP provides significantly better tradeoffs between membership privacy and classification accuracies compared to state-of-the-art MIA defenses. For instance, DMP achieves ~100% accuracy improvement over adversarial regularization for DenseNet trained on CIFAR100, for similar membership privacy (measured using MIA risk): when the MIA risk is 53.7%, adversarially regularized DenseNet is 33.6% accurate, while DMP-trained DenseNet is 65.3% accurate.
We describe a simple and general neural network weight compression approach, in which the network parameters (weights and biases) are represented in a "latent" space, amounting to a reparameterization. This space is equipped with a learned probability model, which is used to impose an entropy penalty on the parameter representation during training, and to compress the representation using a simple arithmetic coder after training. Classification accuracy and model compressibility is maximized jointly, with the bitrate--accuracy trade-off specified by a hyperparameter. We evaluate the method on the MNIST, CIFAR-10 and ImageNet classification benchmarks using six distinct model architectures. Our results show that state-of-the-art model compression can be achieved in a scalable and general way without requiring complex procedures such as multi-stage training.
State-of-the-art results on image recognition tasks are achieved using over-parameterized learning algorithms that (nearly) perfectly fit the training set and are known to fit well even random labels. This tendency to memorize the labels of the training data is not explained by existing theoretical analyses. Memorization of the training data also presents significant privacy risks when the training data contains sensitive personal information and thus it is important to understand whether such memorization is necessary for accurate learning. We provide the first conceptual explanation and a theoretical model for this phenomenon. Specifically, we demonstrate that for natural data distributions memorization of labels is necessary for achieving close-to-optimal generalization error. Crucially, even labels of outliers and noisy labels need to be memorized. The model is motivated and supported by the results of several recent empirical works. In our model, data is sampled from a mixture of subpopulations and our results show that memorization is necessary whenever the distribution of subpopulation frequencies is long-tailed. Image and text data is known to be long-tailed and therefore our results establish a formal link between these empirical phenomena. Our results allow to quantify the cost of limiting memorization in learning and explain the disparate effects that privacy and model compression have on different subgroups.
Residual Networks with convolutional layers are widely used in the field of machine learning. Since they effectively extract features from input data by stacking multiple layers, they can achieve high accuracy in many applications. However, the stacking of many layers raises their computation costs. To address this problem, we propose Network Implosion, it erases multiple layers from Residual Networks without degrading accuracy. Our key idea is to introduce a priority term that identifies the importance of a layer; we can select unimportant layers according to the priority and erase them after the training. In addition, we retrain the networks to avoid critical drops in accuracy after layer erasure. A theoretical assessment reveals that our erasure and retraining scheme can erase layers without accuracy drop, and achieve higher accuracy than is possible with training from scratch. Our experiments show that Network Implosion can, for classification on Cifar-10/100 and ImageNet, reduce the number of layers by 24.00 to 42.86 percent without any drop in accuracy.
Fully convolutional networks (FCNs) have become de facto tool to achieve very high-level performance for many vision and non-vision tasks in general and face recognition in particular. Such high-level accuracies are normally obtained by very deep networks or their ensemble. However, deploying such high performing models to resource constraint devices or real-time applications is challenging. In this paper, we present a novel model compression approach based on student-teacher paradigm for face recognition applications. The proposed approach consists of training teacher FCN at bigger image resolution while student FCNs are trained at lower image resolutions than that of teacher FCN. We explored three different approaches to train student FCNs: knowledge transfer (KT), knowledge distillation (KD) and their combination. Experimental evaluation on LFW and IJB-C datasets demonstrate comparable improvements in accuracies with these approaches. Training low-resolution student FCNs from higher resolution teacher offer fourfold advantage of accelerated training, accelerated inference, reduced memory requirements and improved accuracies. We evaluated all models on IJB-C dataset and achieved state-of-the-art results on this benchmark. The teacher network and some student networks even achieved Top-1 performance on IJB-C dataset. The proposed approach is simple and hardware friendly, thus enables the deployment of high performing face recognition deep models to resource constraint devices.
The training of deep neural networks (DNNs) requires intensive resources both for computation and for storage performance. Thus, DNNs cannot be efficiently applied to mobile phones and embedded devices, which seriously limits their applicability in industry applications. To address this issue, we propose a novel encoding scheme of using {-1,+1} to decompose quantized neural networks (QNNs) into multi-branch binary networks, which can be efficiently implemented by bitwise operations (xnor and bitcount) to achieve model compression, computational acceleration and resource saving. Based on our method, users can easily achieve different encoding precisions arbitrarily according to their requirements and hardware resources. The proposed mechanism is very suitable for the use of FPGA and ASIC in terms of data storage and computation, which provides a feasible idea for smart chips. We validate the effectiveness of our method on both large-scale image classification tasks (e.g., ImageNet) and object detection tasks. In particular, our method with low-bit encoding can still achieve almost the same performance as its full-precision counterparts.
Tensor decomposition is an effective approach to compress over-parameterized neural networks and to enable their deployment on resource-constrained hardware platforms. However, directly applying tensor compression in the training process is a challenging task due to the difficulty of choosing a proper tensor rank. In order to achieve this goal, this paper proposes a Bayesian tensorized neural network. Our Bayesian method performs automatic model compression via an adaptive tensor rank determination. We also present approaches for posterior density calculation and maximum a posteriori (MAP) estimation for the end-to-end training of our tensorized neural network. We provide experimental validation on a fully connected neural network, a CNN and a residual neural network where our work produces $7.4\times$ to $137\times$ more compact neural networks directly from the training.