We propose a software framework based on the ideas of the Learning-Compression (LC) algorithm, that allows a user to compress a neural network or other machine learning model using different compression schemes with minimal effort. Currently, the supported compressions include pruning, quantization, low-rank methods (including automatically learning the layer ranks), and combinations of those, and the user can choose different compression types for different parts of a neural network. The LC algorithm alternates two types of steps until convergence: a learning (L) step, which trains a model on a dataset (using an algorithm such as SGD); and a compression (C) step, which compresses the model parameters (using a compression scheme such as low-rank or quantization). This decoupling of the "machine learning" aspect from the "signal compression" aspect means that changing the model or the compression type amounts to calling the corresponding subroutine in the L or C step, respectively. The library fully supports this by design, which makes it flexible and extensible. This does not come at the expense of performance: the runtime needed to compress a model is comparable to that of training the model in the first place; and the compressed model is competitive in terms of prediction accuracy and compression ratio with other algorithms (which are often specialized for specific models or compression schemes). The library is written in Python and PyTorch and available in Github.
It is important to design compact language models for efficient deployment. We improve upon recent advances in both the language modeling domain and the model-compression domain to construct parameter and computation efficient language models. We use an efficient transformer-based architecture with adaptive embedding and softmax, differentiable non-parametric cache, Hebbian softmax, knowledge distillation, network pruning, and low-bit quantization. In this paper, we provide the winning solution to the NeurIPS 2019 MicroNet Challenge in the language modeling track. Compared to the baseline language model provided by the MicroNet Challenge, our model is 90 times more parameter-efficient and 36 times more computation-efficient while achieving the required test perplexity of 35 on the Wikitext-103 dataset. We hope that this work will aid future research into efficient language models, and we have released our full source code at https://github.com/mit-han-lab/neurips-micronet.
Quantization has been proven to be an effective method for reducing the computing and/or storage cost of DNNs. However, the trade-off between the quantization bitwidth and final accuracy is complex and non-convex, which makes it difficult to be optimized directly. Minimizing direct quantization loss (DQL) of the coefficient data is an effective local optimization method, but previous works often neglect the accurate control of the DQL, resulting in a higher loss of the final DNN model accuracy. In this paper, we propose a novel metric called Vector Loss. Based on this new metric, we develop a new quantization solution called VecQ, which can guarantee minimal direct quantization loss and better model accuracy. In addition, in order to speed up the proposed quantization process during model training, we accelerate the quantization process with a parameterized probability estimation method and template-based derivation calculation. We evaluate our proposed algorithm on MNIST, CIFAR, ImageNet, IMDB movie review and THUCNews text data sets with numerical DNN models. The results demonstrate that our proposed quantization solution is more accurate and effective than the state-of-the-art approaches yet with more flexible bitwidth support. Moreover, the evaluation of our quantized models on Saliency Object Detection (SOD) tasks maintains comparable feature extraction quality with up to 16$\times$ weight size reduction.
Although state-of-the-art (SOTA) CNNs achieve outstanding performance on various tasks, their high computation demand and massive number of parameters make it difficult to deploy these SOTA CNNs onto resource-constrained devices. Previous works on CNN acceleration utilize low-rank approximation of the original convolution layers to reduce computation cost. However, these methods are very difficult to conduct upon sparse models, which limits execution speedup since redundancies within the CNN model are not fully exploited. We argue that kernel granularity decomposition can be conducted with low-rank assumption while exploiting the redundancy within the remaining compact coefficients. Based on this observation, we propose PENNI, a CNN model compression framework that is able to achieve model compactness and hardware efficiency simultaneously by (1) implementing kernel sharing in convolution layers via a small number of basis kernels and (2) alternately adjusting bases and coefficients with sparse constraints. Experiments show that we can prune 97% parameters and 92% FLOPs on ResNet18 CIFAR10 with no accuracy loss, and achieve 44% reduction in run-time memory consumption and a 53% reduction in inference latency.
Second-order information, in the form of Hessian- or Inverse-Hessian-vector products, is a fundamental tool for solving optimization problems. Recently, there has been significant interest in utilizing this information in the context of deep neural networks; however, relatively little is known about the quality of existing approximations in this context. Our work examines this question, identifies issues with existing approaches, and proposes a method called WoodFisher to compute a faithful and efficient estimate of the inverse Hessian. Our main application is to neural network compression, where we build on the classic Optimal Brain Damage/Surgeon framework. We demonstrate that WoodFisher significantly outperforms popular state-of-the-art methods for one-shot pruning. Further, even when iterative, gradual pruning is considered, our method results in a gain in test accuracy over the state-of-the-art approaches, for pruning popular neural networks (like ResNet-50, MobileNetV1) trained on standard image classification datasets such as ImageNet ILSVRC. We examine how our method can be extended to take into account first-order information, as well as illustrate its ability to automatically set layer-wise pruning thresholds and perform compression in the limited-data regime. The code is available at the following link, https://github.com/IST-DASLab/WoodFisher.
We tackle the problem of producing compact models, maximizing their accuracy for a given model size. A standard solution is to train networks with Quantization Aware Training, where the weights are quantized during training and the gradients approximated with the Straight-Through Estimator. In this paper, we extend this approach to work beyond int8 fixed-point quantization with extreme compression methods where the approximations introduced by STE are severe, such as Product Quantization. Our proposal is to only quantize a different random subset of weights during each forward, allowing for unbiased gradients to flow through the other weights. Controlling the amount of noise and its form allows for extreme compression rates while maintaining the performance of the original model. As a result we establish new state-of-the-art compromises between accuracy and model size both in natural language processing and image classification. For example, applying our method to state-of-the-art Transformer and ConvNet architectures, we can achieve 82.5% accuracy on MNLI by compressing RoBERTa to 14MB and 80.0 top-1 accuracy on ImageNet by compressing an EfficientNet-B3 to 3.3MB.
Deep learning networks are being developed in every stage of the MRI workflow and have provided state-of-the-art results. However, this has come at the cost of increased computation requirement and storage. Hence, replacing the networks with compact models at various stages in the MRI workflow can significantly reduce the required storage space and provide considerable speedup. In computer vision, knowledge distillation is a commonly used method for model compression. In our work, we propose a knowledge distillation (KD) framework for the image to image problems in the MRI workflow in order to develop compact, low-parameter models without a significant drop in performance. We propose a combination of the attention-based feature distillation method and imitation loss and demonstrate its effectiveness on the popular MRI reconstruction architecture, DC-CNN. We conduct extensive experiments using Cardiac, Brain, and Knee MRI datasets for 4x, 5x and 8x accelerations. We observed that the student network trained with the assistance of the teacher using our proposed KD framework provided significant improvement over the student network trained without assistance across all the datasets and acceleration factors. Specifically, for the Knee dataset, the student network achieves $65\%$ parameter reduction, 2x faster CPU running time, and 1.5x faster GPU running time compared to the teacher. Furthermore, we compare our attention-based feature distillation method with other feature distillation methods. We also conduct an ablative study to understand the significance of attention-based distillation and imitation loss. We also extend our KD framework for MRI super-resolution and show encouraging results.
Deep neural models in recent years have been successful in almost every field, including extremely complex problem statements. However, these models are huge in size, with millions (and even billions) of parameters, thus demanding more heavy computation power and failing to be deployed on edge devices. Besides, the performance boost is highly dependent on redundant labeled data. To achieve faster speeds and to handle the problems caused by the lack of data, knowledge distillation (KD) has been proposed to transfer information learned from one model to another. KD is often characterized by the so-called `Student-Teacher' (S-T) learning framework and has been broadly applied in model compression and knowledge transfer. This paper is about KD and S-T learning, which are being actively studied in recent years. First, we aim to provide explanations of what KD is and how/why it works. Then, we provide a comprehensive survey on the recent progress of KD methods together with S-T frameworks typically for vision tasks. In general, we consider some fundamental questions that have been driving this research area and thoroughly generalize the research progress and technical details. Additionally, we systematically analyze the research status of KD in vision applications. Finally, we discuss the potentials and open challenges of existing methods and prospect the future directions of KD and S-T learning.
Sparsity-inducing regularization problems are ubiquitous in machine learning applications, ranging from feature selection to model compression. In this paper, we present a novel stochastic method -- Orthant Based Proximal Stochastic Gradient Method (OBProx-SG) -- to solve perhaps the most popular instance, i.e., the l1-regularized problem. The OBProx-SG method contains two steps: (i) a proximal stochastic gradient step to predict a support cover of the solution; and (ii) an orthant step to aggressively enhance the sparsity level via orthant face projection. Compared to the state-of-the-art methods, e.g., Prox-SG, RDA and Prox-SVRG, the OBProx-SG not only converges to the global optimal solutions (in convex scenario) or the stationary points (in non-convex scenario), but also promotes the sparsity of the solutions substantially. Particularly, on a large number of convex problems, OBProx-SG outperforms the existing methods comprehensively in the aspect of sparsity exploration and objective values. Moreover, the experiments on non-convex deep neural networks, e.g., MobileNetV1 and ResNet18, further demonstrate its superiority by achieving the solutions of much higher sparsity without sacrificing generalization accuracy.
The enormous inference cost of deep neural networks can be scaled down by network compression. Pruning is one of the predominant approaches used for deep network compression. However, existing pruning techniques have one or more of the following limitations: 1) Additional energy cost on top of the compute heavy training stage due to pruning and fine-tuning stages, 2) Layer-wise pruning based on the statistics of a particular, ignoring the effect of error propagation in the network, 3) Lack of an efficient estimate for determining the important channels globally, 4) Unstructured pruning requires specialized hardware for effective use. To address all the above issues, we present a simple-yet-effective gradual channel pruning while training methodology using a novel data-driven metric referred to as feature relevance score. The proposed technique gets rid of the additional retraining cycles by pruning the least important channels in a structured fashion at fixed intervals during the actual training phase. Feature relevance scores help in efficiently evaluating the contribution of each channel towards the discriminative power of the network. We demonstrate the effectiveness of the proposed methodology on architectures such as VGG and ResNet using datasets such as CIFAR-10, CIFAR-100 and ImageNet, and successfully achieve significant model compression while trading off less than $1\%$ accuracy. Notably on CIFAR-10 dataset trained on ResNet-110, our approach achieves $2.4\times$ compression and a $56\%$ reduction in FLOPs with an accuracy drop of $0.01\%$ compared to the unpruned network.
We propose a novel algorithm for quantizing continuous latent representations in trained models. Our approach applies to deep probabilistic models, such as variational autoencoders (VAEs), and enables both data and model compression. Unlike current end-to-end neural compression methods that cater the model to a fixed quantization scheme, our algorithm separates model design and training from quantization. Consequently, our algorithm enables "plug-and-play" compression with variable rate-distortion trade-off, using a single trained model. Our algorithm can be seen as a novel extension of arithmetic coding to the continuous domain, and uses adaptive quantization accuracy based on estimates of posterior uncertainty. Our experimental results demonstrate the importance of taking into account posterior uncertainties, and show that image compression with the proposed algorithm outperforms JPEG over a wide range of bit rates using only a single standard VAE. Further experiments on Bayesian neural word embeddings demonstrate the versatility of the proposed method.
In this paper, we propose a novel model compression approach to effectively compress BERT by progressive module replacing. Our approach first divides the original BERT into several modules and builds their compact substitutes. Then, we randomly replace the original modules with their substitutes to train the compact modules to mimic the behavior of the original modules. We progressively increase the probability of replacement through the training. In this way, our approach brings a deeper level of interaction between the original and compact models. Compared to the previous knowledge distillation approaches for BERT compression, our approach does not introduce any additional loss function. Our approach outperforms existing knowledge distillation approaches on GLUE benchmark, showing a new perspective of model compression.
We study network pruning which aims to remove redundant channels/kernels and hence speed up the inference of deep networks. Existing pruning methods either train from scratch with sparsity constraints or minimize the reconstruction error between the feature maps of the pre-trained models and the compressed ones. Both strategies suffer from some limitations: the former kind is computationally expensive and difficult to converge, while the latter kind optimizes the reconstruction error but ignores the discriminative power of channels. In this paper, we propose a simple-yet-effective method called discrimination-aware channel pruning (DCP) to choose the channels that actually contribute to the discriminative power. Note that a channel often consists of a set of kernels. Besides the redundancy in channels, some kernels in a channel may also be redundant and fail to contribute to the discriminative power of the network, resulting in kernel level redundancy. To solve this, we propose a discrimination-aware kernel pruning (DKP) method to further compress deep networks by removing redundant kernels. To prevent DCP/DKP from selecting redundant channels/kernels, we propose a new adaptive stopping condition, which helps to automatically determine the number of selected channels/kernels and often results in more compact models with better performance. Extensive experiments on both image classification and face recognition demonstrate the effectiveness of our methods. For example, on ILSVRC-12, the resultant ResNet-50 model with 30% reduction of channels even outperforms the baseline model by 0.36% in terms of Top-1 accuracy. The pruned MobileNetV1 and MobileNetV2 achieve 1.93x and 1.42x inference acceleration on a mobile device, respectively, with negligible performance degradation. The source code and the pre-trained models are available at https://github.com/SCUT-AILab/DCP.
Quantization is a promising approach for reducing the inference time and memory footprint of neural networks. However, most existing quantization methods require access to the original training dataset for retraining during quantization. This is often not possible for applications with sensitive or proprietary data, e.g., due to privacy and security concerns. Existing zero-shot quantization methods use different heuristics to address this, but they result in poor performance, especially when quantizing to ultra-low precision. Here, we propose ZeroQ , a novel zero-shot quantization framework to address this. ZeroQ enables mixed-precision quantization without any access to the training or validation data. This is achieved by optimizing for a Distilled Dataset, which is engineered to match the statistics of batch normalization across different layers of the network. ZeroQ supports both uniform and mixed-precision quantization. For the latter, we introduce a novel Pareto frontier based method to automatically determine the mixed-precision bit setting for all layers, with no manual search involved. We extensively test our proposed method on a diverse set of models, including ResNet18/50/152, MobileNetV2, ShuffleNet, SqueezeNext, and InceptionV3 on ImageNet, as well as RetinaNet-ResNet50 on the Microsoft COCO dataset. In particular, we show that ZeroQ can achieve 1.71\% higher accuracy on MobileNetV2, as compared to the recently proposed DFQ method. Importantly, ZeroQ has a very low computational overhead, and it can finish the entire quantization process in less than 30s (0.5\% of one epoch training time of ResNet50 on ImageNet). We have open-sourced the ZeroQ framework\footnote{https://github.com/amirgholami/ZeroQ}.
Knowledge Distillation (KD) has made remarkable progress in the last few years and become a popular paradigm for model compression and knowledge transfer. However, almost all existing KD algorithms are data-driven, i.e., relying on a large amount of original training data or alternative data, which is usually unavailable in real-world scenarios. In this paper, we devote ourselves to this challenging problem and propose a novel adversarial distillation mechanism to craft a compact student model without any real-world data. We introduce a model discrepancy to quantificationally measure the difference between student and teacher models and construct an optimizable upper bound. In our work, the student and the teacher jointly act the role of the discriminator to reduce this discrepancy, when a generator adversarially produces some "hard samples" to enlarge it. Extensive experiments demonstrate that the proposed data-free method yields comparable performance to existing data-driven methods. More strikingly, our approach can be directly extended to semantic segmentation, which is more complicated than classification, and our approach achieves state-of-the-art results. Code and pretrained models are available at https://github.com/VainF/Data-Free-Adversarial-Distillation.
The success of convolutional neural networks (CNNs) in various applications is accompanied by a significant increase in computation and parameter storage costs. Recent efforts to reduce these overheads involve pruning and compressing the weights of various layers while at the same time aiming to not sacrifice performance. In this paper, we propose a novel criterion for CNN pruning inspired by neural network interpretability: The most relevant units, i.e. weights or filters, are automatically found using their relevance scores obtained from concepts of explainable AI (XAI). By exploring this idea, we connect the lines of interpretability and model compression research. We show that our proposed method can efficiently prune CNN models in transfer-learning setups in which networks pre-trained on large corpora are adapted to specialized tasks. The method is evaluated on a broad range of computer vision datasets. Notably, our novel criterion is not only competitive or better compared to state-of-the-art pruning criteria when successive retraining is performed, but clearly outperforms these previous criteria in the resource-constrained application scenario in which the data of the task to be transferred to is very scarce and one chooses to refrain from fine-tuning. Our method is able to compress the model iteratively while maintaining or even improving accuracy. At the same time, it has a computational cost in the order of gradient computation and is comparatively simple to apply without the need for tuning hyperparameters for pruning.
While state-of-the-art NLP explainability (XAI) methods focus on explaining per-sample decisions in supervised end or probing tasks, this is insufficient to explain and quantify model knowledge transfer during (un-)supervised training. Thus, for TX-Ray, we modify the established computer vision explainability principle of 'visualizing preferred inputs of neurons' to make it usable transfer analysis and NLP. This allows one to analyze, track and quantify how self- or supervised NLP models first build knowledge abstractions in pretraining (1), and then transfer these abstractions to a new domain (2), or adapt them during supervised fine-tuning (3). TX-Ray expresses neurons as feature preference distributions to quantify fine-grained knowledge transfer or adaptation and guide human analysis. We find that, similar to Lottery Ticket based pruning, TX-Ray based pruning can improve test set generalization and that it can reveal how early stages of self-supervision automatically learn linguistic abstractions like parts-of-speech.
Model compression has been widely adopted to obtain light-weighted deep neural networks. Most prevalent methods, however, require fine-tuning with sufficient training data to ensure accuracy, which could be challenged by privacy and security issues. As a compromise between privacy and performance, in this paper we investigate few shot network compression: given few samples per class, how can we effectively compress the network with negligible performance drop? The core challenge of few shot network compression lies in high estimation errors from the original network during inference, since the compressed network can easily over-fits on the few training instances. The estimation errors could propagate and accumulate layer-wisely and finally deteriorate the network output. To address the problem, we propose cross distillation, a novel layer-wise knowledge distillation approach. By interweaving hidden layers of teacher and student network, layer-wisely accumulated estimation errors can be effectively reduced.The proposed method offers a general framework compatible with prevalent network compression techniques such as pruning. Extensive experiments on benchmark datasets demonstrate that cross distillation can significantly improve the student network's accuracy when only a few training instances are available.
Compressed communication, in the form of sparsification or quantization of stochastic gradients, is employed to reduce communication costs in distributed data-parallel training of deep neural networks. However, there exists a discrepancy between theory and practice: while theoretical analysis of most existing compression methods assumes compression is applied to the gradients of the entire model, many practical implementations operate individually on the gradients of each layer of the model. In this paper, we prove that layer-wise compression is, in theory, better, because the convergence rate is upper bounded by that of entire-model compression for a wide range of biased and unbiased compression methods. However, despite the theoretical bound, our experimental study of six well-known methods shows that convergence, in practice, may or may not be better, depending on the actual trained model and compression ratio. Our findings suggest that it would be advantageous for deep learning frameworks to include support for both layer-wise and entire-model compression.
Despite the success of Deep Learning (DL), the deployment of modern DL models requiring large computational power poses a significant problem for resource-constrained systems. This necessitates building compact networks that reduce computations while preserving performance. Traditional Knowledge Distillation (KD) methods that transfer knowledge from teacher to student (a) use a single-stage and (b) require the whole data set while distilling the knowledge to the student. In this work, we propose a new method called Stagewise Knowledge Distillation (SKD) which builds on traditional KD methods by progressive stagewise training to leverage the knowledge gained from the teacher, resulting in data-efficient distillation process. We evaluate our method on classification and semantic segmentation tasks. We show, across the tested tasks, significant performance gains even with a fraction of the data used in distillation, without compromising on the metric. We also compare our method with existing KD techniques and show that SKD outperforms them. Moreover, our method can be viewed as a generalized model compression technique that complements other model compression methods such as quantization or pruning.