PyTorchKR Papers

최신 논문

총 4,618편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2005.02177

CDC: Classification Driven Compression for Bandwidth Efficient Edge-Cloud Collaborative Deep Learning

The emerging edge-cloud collaborative Deep Learning (DL) paradigm aims at improving the performance of practical DL implementations in terms of cloud bandwidth consumption, response latency, and data privacy preservation. Focusing on bandwidth efficient edge-cloud collaborative training of DNN-based classifiers, we present CDC, a Classification Driven Compression framework that reduces bandwidth consumption while preserving classification accuracy of edge-cloud collaborative DL. Specifically, to reduce bandwidth consumption, for resource-limited edge servers, we develop a lightweight autoencoder with a classification guidance for compression with classification driven feature preservation, which allows edges to only upload the latent code of raw data for accurate global training on the Cloud. Additionally, we design an adjustable quantization scheme adaptively pursuing the tradeoff between bandwidth consumption and classification accuracy under different network conditions, where only fine-tuning is required for rapid compression ratio adjustment. Results of extensive experiments demonstrate that, compared with DNN training with raw data, CDC consumes 14.9 times less bandwidth with an accuracy loss no more than 1.06%, and compared with DNN training with data compressed by AE without guidance, CDC introduces at least 100% lower accuracy loss.

arXiv:2005.01091공식 코드 · Papers with Code

A Little Bit More: Bitplane-Wise Bit-Depth Recovery

Imaging sensors digitize incoming scene light at a dynamic range of 10--12 bits (i.e., 1024--4096 tonal values). The sensor image is then processed onboard the camera and finally quantized to only 8 bits (i.e., 256 tonal values) to conform to prevailing encoding standards. There are a number of important applications, such as high-bit-depth displays and photo editing, where it is beneficial to recover the lost bit depth. Deep neural networks are effective at this bit-depth reconstruction task. Given the quantized low-bit-depth image as input, existing deep learning methods employ a single-shot approach that attempts to either (1) directly estimate the high-bit-depth image, or (2) directly estimate the residual between the high- and low-bit-depth images. In contrast, we propose a training and inference strategy that recovers the residual image bitplane-by-bitplane. Our bitplane-wise learning framework has the advantage of allowing for multiple levels of supervision during training and is able to obtain state-of-the-art results using a simple network architecture. We test our proposed method extensively on several image datasets and demonstrate an improvement from 0.5dB to 2.3dB PSNR over prior methods depending on the quantization level.

arXiv:2004.13770

Streamlining Tensor and Network Pruning in PyTorch

In order to contrast the explosion in size of state-of-the-art machine learning models that can be attributed to the empirical advantages of over-parametrization, and due to the necessity of deploying fast, sustainable, and private on-device models on resource-constrained devices, the community has focused on techniques such as pruning, quantization, and distillation as central strategies for model compression. Towards the goal of facilitating the adoption of a common interface for neural network pruning in PyTorch, this contribution describes the recent addition of the PyTorch torch.nn.utils.prune module, which provides shared, open source pruning functionalities to lower the technical implementation barrier to reducing model size and capacity before, during, and/or after training. We present the module's user interface, elucidate implementation details, illustrate example usage, and suggest ways to extend the contributed functionalities to new pruning methods.

arXiv:1812.07106

E-RNN: Design Optimization for Efficient Recurrent Neural Networks in FPGAs

Recurrent Neural Networks (RNNs) are becoming increasingly important for time series-related applications which require efficient and real-time implementations. The two major types are Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks. It is a challenging task to have real-time, efficient, and accurate hardware RNN implementations because of the high sensitivity to imprecision accumulation and the requirement of special activation function implementations. A key limitation of the prior works is the lack of a systematic design optimization framework of RNN model and hardware implementations, especially when the block size (or compression ratio) should be jointly optimized with RNN type, layer size, etc. In this paper, we adopt the block-circulant matrix-based framework, and present the Efficient RNN (E-RNN) framework for FPGA implementations of the Automatic Speech Recognition (ASR) application. The overall goal is to improve performance/energy efficiency under accuracy requirement. We use the alternating direction method of multipliers (ADMM) technique for more accurate block-circulant training, and present two design explorations providing guidance on block size and reducing RNN training trials. Based on the two observations, we decompose E-RNN in two phases: Phase I on determining RNN model to reduce computation and storage subject to accuracy requirement, and Phase II on hardware implementations given RNN model, including processing element design/optimization, quantization, activation implementation, etc. Experimental results on actual FPGA deployments show that E-RNN achieves a maximum energy efficiency improvement of 37.4$\times$ compared with ESE, and more than 2$\times$ compared with C-LSTM, under the same accuracy.

arXiv:2004.14180

Quantized Adam with Error Feedback

In this paper, we present a distributed variant of adaptive stochastic gradient method for training deep neural networks in the parameter-server model. To reduce the communication cost among the workers and server, we incorporate two types of quantization schemes, i.e., gradient quantization and weight quantization, into the proposed distributed Adam. Besides, to reduce the bias introduced by quantization operations, we propose an error-feedback technique to compensate for the quantized gradient. Theoretically, in the stochastic nonconvex setting, we show that the distributed adaptive gradient method with gradient quantization and error-feedback converges to the first-order stationary point, and that the distributed adaptive gradient method with weight quantization and error-feedback converges to the point related to the quantized level under both the single-worker and multi-worker modes. At last, we apply the proposed distributed adaptive gradient methods to train deep neural networks. Experimental results demonstrate the efficacy of our methods.

arXiv:2004.14287

General Purpose Text Embeddings from Pre-trained Language Models for Scalable Inference

The state of the art on many NLP tasks is currently achieved by large pre-trained language models, which require a considerable amount of computation. We explore a setting where many different predictions are made on a single piece of text. In that case, some of the computational cost during inference can be amortized over the different tasks using a shared text encoder. We compare approaches for training such an encoder and show that encoders pre-trained over multiple tasks generalize well to unseen tasks. We also compare ways of extracting fixed- and limited-size representations from this encoder, including different ways of pooling features extracted from multiple layers or positions. Our best approach compares favorably to knowledge distillation, achieving higher accuracy and lower computational cost once the system is handling around 7 tasks. Further, we show that through binary quantization, we can reduce the size of the extracted representations by a factor of 16 making it feasible to store them for later use. The resulting method offers a compelling solution for using large-scale pre-trained models at a fraction of the computational cost when multiple tasks are performed on the same text.

arXiv:2004.11506

Automatic low-bit hybrid quantization of neural networks through meta learning

Model quantization is a widely used technique to compress and accelerate deep neural network (DNN) inference, especially when deploying to edge or IoT devices with limited computation capacity and power consumption budget. The uniform bit width quantization across all the layers is usually sub-optimal and the exploration of hybrid quantization for different layers is vital for efficient deep compression. In this paper, we employ the meta learning method to automatically realize low-bit hybrid quantization of neural networks. A MetaQuantNet, together with a Quantization function, are trained to generate the quantized weights for the target DNN. Then, we apply a genetic algorithm to search the best hybrid quantization policy that meets compression constraints. With the best searched quantization policy, we subsequently retrain or finetune to further improve the performance of the quantized target network. Extensive experiments demonstrate the performance of searched hybrid quantization scheme surpass that of uniform bitwidth counterpart. Compared to the existing reinforcement learning (RL) based hybrid quantization search approach that relies on tedious explorations, our meta learning approach is more efficient and effective for any compression requirements since the MetaQuantNet only needs be trained once.

arXiv:2004.11886공식 코드 · Papers with Code

Lite Transformer with Long-Short Range Attention

Transformer has become ubiquitous in natural language processing (e.g., machine translation, question answering); however, it requires enormous amount of computations to achieve high performance, which makes it not suitable for mobile applications that are tightly constrained by the hardware resources and battery. In this paper, we present an efficient mobile NLP architecture, Lite Transformer to facilitate deploying mobile NLP applications on edge devices. The key primitive is the Long-Short Range Attention (LSRA), where one group of heads specializes in the local context modeling (by convolution) while another group specializes in the long-distance relationship modeling (by attention). Such specialization brings consistent improvement over the vanilla transformer on three well-established language tasks: machine translation, abstractive summarization, and language modeling. Under constrained resources (500M/100M MACs), Lite Transformer outperforms transformer on WMT'14 English-French by 1.2/1.7 BLEU, respectively. Lite Transformer reduces the computation of transformer base model by 2.5x with 0.3 BLEU score degradation. Combining with pruning and quantization, we further compressed the model size of Lite Transformer by 18.2x. For language modeling, Lite Transformer achieves 1.8 lower perplexity than the transformer at around 500M MACs. Notably, Lite Transformer outperforms the AutoML-based Evolved Transformer by 0.5 higher BLEU for the mobile NLP setting without the costly architecture search that requires more than 250 GPU years. Code has been made available at https://github.com/mit-han-lab/lite-transformer.

arXiv:2004.11783

Quantization of Deep Neural Networks for Accumulator-constrained Processors

We introduce an Artificial Neural Network (ANN) quantization methodology for platforms without wide accumulation registers. This enables fixed-point model deployment on embedded compute platforms that are not specifically designed for large kernel computations (i.e. accumulator-constrained processors). We formulate the quantization problem as a function of accumulator size, and aim to maximize the model accuracy by maximizing bit width of input data and weights. To reduce the number of configurations to consider, only solutions that fully utilize the available accumulator bits are being tested. We demonstrate that 16-bit accumulators are able to obtain a classification accuracy within 1\% of the floating-point baselines on the CIFAR-10 and ILSVRC2012 image classification benchmarks. Additionally, a near-optimal $2\times$ speedup is obtained on an ARM processor, by exploiting 16-bit accumulators for image classification on the All-CNN-C and AlexNet networks.

arXiv:2004.11233

QUANOS- Adversarial Noise Sensitivity Driven Hybrid Quantization of Neural Networks

Deep Neural Networks (DNNs) have been shown to be vulnerable to adversarial attacks, wherein, a model gets fooled by applying slight perturbations on the input. With the advent of Internet-of-Things and the necessity to enable intelligence in embedded devices, low-power and secure hardware implementation of DNNs is vital. In this paper, we investigate the use of quantization to potentially resist adversarial attacks. Several recent studies have reported remarkable results in reducing the energy requirement of a DNN through quantization. However, no prior work has considered the relationship between adversarial sensitivity of a DNN and its effect on quantization. We propose QUANOS- a framework that performs layer-specific hybrid quantization based on Adversarial Noise Sensitivity (ANS). We identify a novel noise stability metric (ANS) for DNNs, i.e., the sensitivity of each layer's computation to adversarial noise. ANS allows for a principled way of determining optimal bit-width per layer that incurs adversarial robustness as well as energy-efficiency with minimal loss in accuracy. Essentially, QUANOS assigns layer significance based on its contribution to adversarial perturbation and accordingly scales the precision of the layers. A key advantage of QUANOS is that it does not rely on a pre-trained model and can be applied in the initial stages of training. We evaluate the benefits of QUANOS on precision scalable Multiply and Accumulate (MAC) hardware architectures with data gating and subword parallelism capabilities. Our experiments on CIFAR10, CIFAR100 datasets show that QUANOS outperforms homogenously quantized 8-bit precision baseline in terms of adversarial robustness (3%-4% higher) while yielding improved compression (>5x) and energy savings (>2x) at iso-accuracy.

arXiv:2004.10568

Up or Down? Adaptive Rounding for Post-Training Quantization

When quantizing neural networks, assigning each floating-point weight to its nearest fixed-point value is the predominant approach. We find that, perhaps surprisingly, this is not the best we can do. In this paper, we propose AdaRound, a better weight-rounding mechanism for post-training quantization that adapts to the data and the task loss. AdaRound is fast, does not require fine-tuning of the network, and only uses a small amount of unlabelled data. We start by theoretically analyzing the rounding problem for a pre-trained neural network. By approximating the task loss with a Taylor series expansion, the rounding task is posed as a quadratic unconstrained binary optimization problem. We simplify this to a layer-wise local loss and propose to optimize this loss with a soft relaxation. AdaRound not only outperforms rounding-to-nearest by a significant margin but also establishes a new state-of-the-art for post-training quantization on several networks and tasks. Without fine-tuning, we can quantize the weights of Resnet18 and Resnet50 to 4 bits while staying within an accuracy loss of 1%.

arXiv:2004.09691

A Data and Compute Efficient Design for Limited-Resources Deep Learning

Thanks to their improved data efficiency, equivariant neural networks have gained increased interest in the deep learning community. They have been successfully applied in the medical domain where symmetries in the data can be effectively exploited to build more accurate and robust models. To be able to reach a much larger body of patients, mobile, on-device implementations of deep learning solutions have been developed for medical applications. However, equivariant models are commonly implemented using large and computationally expensive architectures, not suitable to run on mobile devices. In this work, we design and test an equivariant version of MobileNetV2 and further optimize it with model quantization to enable more efficient inference. We achieve close-to state of the art performance on the Patch Camelyon (PCam) medical dataset while being more computationally efficient.

arXiv:2004.09602공식 코드 · Papers with Code

Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation

Quantization techniques can reduce the size of Deep Neural Networks and improve inference latency and throughput by taking advantage of high throughput integer instructions. In this paper we review the mathematical aspects of quantization parameters and evaluate their choices on a wide range of neural network models for different application domains, including vision, speech, and language. We focus on quantization techniques that are amenable to acceleration by processors with high-throughput integer math pipelines. We also present a workflow for 8-bit quantization that is able to maintain accuracy within 1% of the floating-point baseline on all networks studied, including models that are more difficult to quantize, such as MobileNets and BERT-large.

arXiv:2004.09576

LSQ+: Improving low-bit quantization through learnable offsets and better initialization

Unlike ReLU, newer activation functions (like Swish, H-swish, Mish) that are frequently employed in popular efficient architectures can also result in negative activation values, with skewed positive and negative ranges. Typical learnable quantization schemes [PACT, LSQ] assume unsigned quantization for activations and quantize all negative activations to zero which leads to significant loss in performance. Naively using signed quantization to accommodate these negative values requires an extra sign bit which is expensive for low-bit (2-, 3-, 4-bit) quantization. To solve this problem, we propose LSQ+, a natural extension of LSQ, wherein we introduce a general asymmetric quantization scheme with trainable scale and offset parameters that can learn to accommodate the negative activations. Gradient-based learnable quantization schemes also commonly suffer from high instability or variance in the final training performance, hence requiring a great deal of hyper-parameter tuning to reach a satisfactory performance. LSQ+ alleviates this problem by using an MSE-based initialization scheme for the quantization parameters. We show that this initialization leads to significantly lower variance in final performance across multiple training runs. Overall, LSQ+ shows state-of-the-art results for EfficientNet and MixNet and also significantly outperforms LSQ for low-bit quantization of neural nets with Swish activations (e.g.: 1.8% gain with W4A4 quantization and upto 5.6% gain with W2A2 quantization of EfficientNet-B0 on ImageNet dataset). To the best of our knowledge, ours is the first work to quantize such architectures to extremely low bit-widths.

arXiv:2004.09320공식 코드 · Papers with Code

Quantization Guided JPEG Artifact Correction

The JPEG image compression algorithm is the most popular method of image compression because of its ability for large compression ratios. However, to achieve such high compression, information is lost. For aggressive quantization settings, this leads to a noticeable reduction in image quality. Artifact correction has been studied in the context of deep neural networks for some time, but the current state-of-the-art methods require a different model to be trained for each quality setting, greatly limiting their practical application. We solve this problem by creating a novel architecture which is parameterized by the JPEG files quantization matrix. This allows our single model to achieve state-of-the-art performance over models trained for specific quality settings.

arXiv:2004.08906

HCM: Hardware-Aware Complexity Metric for Neural Network Architectures

Convolutional Neural Networks (CNNs) have become common in many fields including computer vision, speech recognition, and natural language processing. Although CNN hardware accelerators are already included as part of many SoC architectures, the task of achieving high accuracy on resource-restricted devices is still considered challenging, mainly due to the vast number of design parameters that need to be balanced to achieve an efficient solution. Quantization techniques, when applied to the network parameters, lead to a reduction of power and area and may also change the ratio between communication and computation. As a result, some algorithmic solutions may suffer from lack of memory bandwidth or computational resources and fail to achieve the expected performance due to hardware constraints. Thus, the system designer and the micro-architect need to understand at early development stages the impact of their high-level decisions (e.g., the architecture of the CNN and the amount of bits used to represent its parameters) on the final product (e.g., the expected power saving, area, and accuracy). Unfortunately, existing tools fall short of supporting such decisions. This paper introduces a hardware-aware complexity metric that aims to assist the system designer of the neural network architectures, through the entire project lifetime (especially at its early stages) by predicting the impact of architectural and micro-architectural decisions on the final product. We demonstrate how the proposed metric can help evaluate different design alternatives of neural network models on resource-restricted devices such as real-time embedded systems, and to avoid making design mistakes at early stages.

arXiv:2004.08287

Deep Neural Network for Respiratory Sound Classification in Wearable Devices Enabled by Patient Specific Model Tuning

The primary objective of this paper is to build classification models and strategies to identify breathing sound anomalies (wheeze, crackle) for automated diagnosis of respiratory and pulmonary diseases. In this work we propose a deep CNN-RNN model that classifies respiratory sounds based on Mel-spectrograms. We also implement a patient specific model tuning strategy that first screens respiratory patients and then builds patient specific classification models using limited patient data for reliable anomaly detection. Moreover, we devise a local log quantization strategy for model weights to reduce the memory footprint for deployment in memory constrained systems such as wearable devices. The proposed hybrid CNN-RNN model achieves a score of 66.31% on four-class classification of breathing cycles for ICBHI'17 scientific challenge respiratory sound database. When the model is re-trained with patient specific data, it produces a score of 71.81% for leave-one-out validation. The proposed weight quantization technique achieves ~4X reduction in total memory cost without loss of performance. The main contribution of the paper is as follows: Firstly, the proposed model is able to achieve state of the art score on the ICBHI'17 dataset. Secondly, deep learning models are shown to successfully learn domain specific knowledge when pre-trained with breathing data and produce significantly superior performance compared to generalized models. Finally, local log quantization of trained weights is shown to be able to reduce the memory requirement significantly. This type of patient-specific re-training strategy can be very useful in developing reliable long-term automated patient monitoring systems particularly in wearable healthcare solutions.

arXiv:2004.07456

Single upper limb pose estimation method based on improved stacked hourglass network

At present, most high-accuracy single-person pose estimation methods have high computational complexity and insufficient real-time performance due to the complex structure of the network model. However, a single-person pose estimation method with high real-time performance also needs to improve its accuracy due to the simple structure of the network model. It is currently difficult to achieve both high accuracy and real-time performance in single-person pose estimation. For use in human-machine cooperative operations, this paper proposes a single-person upper limb pose estimation method based on an end-to-end approach for accurate and real-time limb pose estimation. Using the stacked hourglass network model, a single-person upper limb skeleton key point detection model was designed.Deconvolution was employed to replace the up-sampling operation of the hourglass module in the original model, solving the problem of rough feature maps. Integral regression was used to calculate the position coordinates of key points of the skeleton, reducing quantization errors and calculations. Experiments showed that the developed single-person upper limb skeleton key point detection model achieves high accuracy and that the pose estimation method based on the end-to-end approach provides high accuracy and real-time performance.

arXiv:2004.07116

Q-CapsNets: A Specialized Framework for Quantizing Capsule Networks

Capsule Networks (CapsNets), recently proposed by the Google Brain team, have superior learning capabilities in machine learning tasks, like image classification, compared to the traditional CNNs. However, CapsNets require extremely intense computations and are difficult to be deployed in their original form at the resource-constrained edge devices. This paper makes the first attempt to quantize CapsNet models, to enable their efficient edge implementations, by developing a specialized quantization framework for CapsNets. We evaluate our framework for several benchmarks. On a deep CapsNet model for the CIFAR10 dataset, the framework reduces the memory footprint by 6.2x, with only 0.15% accuracy loss. We will open-source our framework at https://git.io/JvDIF in August 2020.

arXiv:2004.06560

Breaking the waves: asymmetric random periodic features for low-bitrate kernel machines

Many signal processing and machine learning applications are built from evaluating a kernel on pairs of signals, e.g. to assess the similarity of an incoming query to a database of known signals. This nonlinear evaluation can be simplified to a linear inner product of the random Fourier features of those signals: random projections followed by a periodic map, the complex exponential. It is known that a simple quantization of those features (corresponding to replacing the complex exponential by a different periodic map that takes binary values, which is appealing for their transmission and storage), distorts the approximated kernel, which may be undesirable in practice. Our take-home message is that when the features of only one of the two signals are quantized, the original kernel is recovered without distortion; its practical interest appears in several cases where the kernel evaluations are asymmetric by nature, such as a client-server scheme. Concretely, we introduce the general framework of asymmetric random periodic features, where the two signals of interest are observed through random periodic features: random projections followed by a general periodic map, which is allowed to be different for both signals. We derive the influence of those periodic maps on the approximated kernel, and prove uniform probabilistic error bounds holding for all signal pairs from an infinite low-complexity set. Interestingly, our results allow the periodic maps to be discontinuous, thanks to a new mathematical tool, i.e. the mean Lipschitz smoothness. We then apply this generic framework to semi-quantized kernel machines (where only one signal has quantized features and the other has classical random Fourier features), for which we show theoretically that the approximated kernel remains unchanged (with the associated error bound), and confirm the power of the approach with numerical simulations.