Post-training Neural Network (NN) model compression is an attractive approach for deploying large, memory-consuming models on devices with limited memory resources. In this study, we investigate the rate-distortion tradeoff for NN model compression. First, we suggest a Rotation-Invariant Quantization (RIQ) technique that utilizes a single parameter to quantize the entire NN model, yielding a different rate at each layer, i.e., mixed-precision quantization. Then, we prove that our rotation-invariant approach is optimal in terms of compression. We rigorously evaluate RIQ and demonstrate its capabilities on various models and tasks. For example, RIQ facilitates $\times 19.4$ and $\times 52.9$ compression ratios on pre-trained VGG dense and pruned models, respectively, with $<0.4\%$ accuracy degradation. Code is available in \href{https://github.com/ehaleva/RIQ}{github.com/ehaleva/RIQ}.
Machine learning (ML) algorithms are increasingly being integrated into embedded and IoT systems that surround us, and they are vulnerable to adversarial attacks. The deployment of these ML algorithms on resource-limited embedded platforms also requires the use of model compression techniques. The impact of such model compression techniques on adversarial robustness in ML is an important and emerging area of research. This article provides an overview of the landscape of adversarial attacks and ML model compression techniques relevant to embedded systems. We then describe efforts that seek to understand the relationship between adversarial attacks and ML model compression before discussing open problems in this area.
Knowledge distillation has widely been used for model compression and domain adaptation for speech applications. In the presence of multiple teachers, knowledge can easily be transferred to the student by averaging the models output. However, previous research shows that the student do not adapt well with such combination. This paper propose to use an elitist sampling strategy at the output of ensemble teacher models to select the best-decoded utterance generated by completely out-of-domain teacher models for generalizing unseen domain. The teacher models are trained on AMI, LibriSpeech and WSJ while the student is adapted for the Switchboard data. The results show that with the selection strategy based on the individual models posteriors the student model achieves a better WER compared to all the teachers and baselines with a minimum absolute improvement of about 8.4 percent. Furthermore, an insights on the model adaptation with out-of-domain data has also been studied via correlation analysis.
Self-supervised speech representation learning (SSL) has shown to be effective in various downstream tasks, but SSL models are usually large and slow. Model compression techniques such as pruning aim to reduce the model size and computation without degradation in accuracy. Prior studies focus on the pruning of Transformers; however, speech models not only utilize a stack of Transformer blocks, but also combine a frontend network based on multiple convolutional layers for low-level feature representation learning. This frontend has a small size but a heavy computational cost. In this work, we propose three task-specific structured pruning methods to deal with such heterogeneous networks. Experiments on LibriSpeech and SLURP show that the proposed method is more accurate than the original wav2vec2-base with 10% to 30% less computation, and is able to reduce the computation by 40% to 50% without any degradation.
Deep models are susceptible to learning spurious correlations, even during the post-processing. We take a closer look at the knowledge distillation -- a popular post-processing technique for model compression -- and find that distilling with biased training data gives rise to a biased student, even when the teacher is debiased. To address this issue, we propose a simple knowledge distillation algorithm, coined DeTT (Debiasing by Teacher Transplanting). Inspired by a recent observation that the last neural net layer plays an overwhelmingly important role in debiasing, DeTT directly transplants the teacher's last layer to the student. Remaining layers are distilled by matching the feature map outputs of the student and the teacher, where the samples are reweighted to mitigate the dataset bias. Importantly, DeTT does not rely on the availability of extensive annotations on the bias-related attribute, which is typically not available during the post-processing phase. Throughout our experiments, DeTT successfully debiases the student model, consistently outperforming the baselines in terms of the worst-group accuracy.
With the growth of neural network size, model compression has attracted increasing interest in recent research. As one of the most common techniques, pruning has been studied for a long time. By exploiting the structured sparsity of the neural network, existing methods can prune neurons instead of individual weights. However, in most existing pruning methods, surviving neurons are randomly connected in the neural network without any structure, and the non-zero weights within each neuron are also randomly distributed. Such irregular sparse structure can cause very high control overhead and irregular memory access for the hardware and even increase the neural network computational complexity. In this paper, we propose a three-layer hierarchical prior to promote a more regular sparse structure during pruning. The proposed three-layer hierarchical prior can achieve per-neuron weight-level structured sparsity and neuron-level structured sparsity. We derive an efficient Turbo-variational Bayesian inferencing (Turbo-VBI) algorithm to solve the resulting model compression problem with the proposed prior. The proposed Turbo-VBI algorithm has low complexity and can support more general priors than existing model compression algorithms. Simulation results show that our proposed algorithm can promote a more regular structure in the pruned neural networks while achieving even better performance in terms of compression rate and inferencing accuracy compared with the baselines.
Knowledge distillation has been shown to be a powerful model compression approach to facilitate the deployment of pre-trained language models in practice. This paper focuses on task-agnostic distillation. It produces a compact pre-trained model that can be easily fine-tuned on various tasks with small computational costs and memory footprints. Despite the practical benefits, task-agnostic distillation is challenging. Since the teacher model has a significantly larger capacity and stronger representation power than the student model, it is very difficult for the student to produce predictions that match the teacher's over a massive amount of open-domain training data. Such a large prediction discrepancy often diminishes the benefits of knowledge distillation. To address this challenge, we propose Homotopic Distillation (HomoDistil), a novel task-agnostic distillation approach equipped with iterative pruning. Specifically, we initialize the student model from the teacher model, and iteratively prune the student's neurons until the target width is reached. Such an approach maintains a small discrepancy between the teacher's and student's predictions throughout the distillation process, which ensures the effectiveness of knowledge transfer. Extensive experiments demonstrate that HomoDistil achieves significant improvements on existing baselines.
The union of Edge Computing (EC) and Artificial Intelligence (AI) has brought forward the Edge AI concept to provide intelligent solutions close to the end-user environment, for privacy preservation, low latency to real-time performance, and resource optimization. Machine Learning (ML), as the most advanced branch of AI in the past few years, has shown encouraging results and applications in the edge environment. Nevertheless, edge-powered ML solutions are more complex to realize due to the joint constraints from both edge computing and AI domains, and the corresponding solutions are expected to be efficient and adapted in technologies such as data processing, model compression, distributed inference, and advanced learning paradigms for Edge ML requirements. Despite the fact that a great deal of the attention garnered by Edge ML is gained in both the academic and industrial communities, we noticed the lack of a complete survey on existing Edge ML technologies to provide a common understanding of this concept. To tackle this, this paper aims at providing a comprehensive taxonomy and a systematic review of Edge ML techniques, focusing on the soft computing aspects of existing paradigms and techniques. We start by identifying the Edge ML requirements driven by the joint constraints. We then extensively survey more than twenty paradigms and techniques along with their representative work, covering two main parts: edge inference, and edge learning. In particular, we analyze how each technique fits into Edge ML by meeting a subset of the identified requirements. We also summarize Edge ML frameworks and open issues to shed light on future directions for Edge ML.
Model compression is instrumental in optimizing deep neural network inference on resource-constrained hardware. The prevailing methods for network compression, namely quantization and pruning, have been shown to enhance efficiency at the cost of performance. Determining the most effective quantization and pruning strategies for individual layers and parameters remains a challenging problem, often requiring computationally expensive and ad hoc numerical optimization techniques. This paper introduces FITCompress, a novel method integrating layer-wise mixed-precision quantization and unstructured pruning using a unified heuristic approach. By leveraging the Fisher Information Metric and path planning through compression space, FITCompress optimally selects a combination of pruning mask and mixed-precision quantization configuration for a given pre-trained model and compression constraint. Experiments on computer vision and natural language processing benchmarks demonstrate that our proposed approach achieves a superior compression-performance trade-off compared to existing state-of-the-art methods. FITCompress stands out for its principled derivation, making it versatile across tasks and network architectures, and represents a step towards achieving optimal compression for neural networks.
Knowledge distillation is an effective method for model compression. However, it is still a challenging topic to apply knowledge distillation to detection tasks. There are two key points resulting in poor distillation performance for detection tasks. One is the serious imbalance between foreground and background features, another one is that small object lacks enough feature representation. To solve the above issues, we propose a new distillation method named dual relation knowledge distillation (DRKD), including pixel-wise relation distillation and instance-wise relation distillation. The pixel-wise relation distillation embeds pixel-wise features in the graph space and applies graph convolution to capture the global pixel relation. By distilling the global pixel relation, the student detector can learn the relation between foreground and background features, and avoid the difficulty of distilling features directly for the feature imbalance issue. Besides, we find that instance-wise relation supplements valuable knowledge beyond independent features for small objects. Thus, the instance-wise relation distillation is designed, which calculates the similarity of different instances to obtain a relation matrix. More importantly, a relation filter module is designed to highlight valuable instance relations. The proposed dual relation knowledge distillation is general and can be easily applied for both one-stage and two-stage detectors. Our method achieves state-of-the-art performance, which improves Faster R-CNN based on ResNet50 from 38.4% to 41.6% mAP and improves RetinaNet based on ResNet50 from 37.4% to 40.3% mAP on COCO 2017.
Deep Neural Network (DNN) Inference in Edge Computing, often called Edge Intelligence, requires solutions to insure that sensitive data confidentiality and intellectual property are not revealed in the process. Privacy-preserving Edge Intelligence is only emerging, despite the growing prevalence of Edge Computing as a context of Machine-Learning-as-a-Service. Solutions are yet to be applied, and possibly adapted, to state-of-the-art DNNs. This position paper provides an original assessment of the compatibility of existing techniques for privacy-preserving DNN Inference with the characteristics of an Edge Computing setup, highlighting the appropriateness of secret sharing in this context. We then address the future role of model compression methods in the research towards secret sharing on DNNs with state-of-the-art performance.
In Natural Language Processing (NLP), Transformers have already revolutionized the field by utilizing an attention-based encoder-decoder model. Recently, some pioneering works have employed Transformer-like architectures in Computer Vision (CV) and they have reported outstanding performance of these architectures in tasks such as image classification, object detection, and semantic segmentation. Vision Transformers (ViTs) have demonstrated impressive performance improvements over Convolutional Neural Networks (CNNs) due to their competitive modelling capabilities. However, these architectures demand massive computational resources which makes these models difficult to be deployed in the resource-constrained applications. Many solutions have been developed to combat this issue, such as compressive transformers and compression functions such as dilated convolution, min-max pooling, 1D convolution, etc. Model compression has recently attracted considerable research attention as a potential remedy. A number of model compression methods have been proposed in the literature such as weight quantization, weight multiplexing, pruning and Knowledge Distillation (KD). However, techniques like weight quantization, pruning and weight multiplexing typically involve complex pipelines for performing the compression. KD has been found to be a simple and much effective model compression technique that allows a relatively simple model to perform tasks almost as accurately as a complex model. This paper discusses various approaches based upon KD for effective compression of ViT models. The paper elucidates the role played by KD in reducing the computational and memory requirements of these models. The paper also presents the various challenges faced by ViTs that are yet to be resolved.
Deep neural networks have seen enormous success in various real-world applications. Beyond their predictions as point estimates, increasing attention has been focused on quantifying the uncertainty of their predictions. In this review, we show that the uncertainty of deep neural networks is not only important in a sense of interpretability and transparency, but also crucial in further advancing their performance, particularly in learning systems seeking robustness and efficiency. We will generalize the definition of the uncertainty of deep neural networks to any number or vector that is associated with an input or an input-label pair, and catalog existing methods on ``mining'' such uncertainty from a deep model. We will include those methods from the classic field of uncertainty quantification as well as those methods that are specific to deep neural networks. We then show a wide spectrum of applications of such generalized uncertainty in realistic learning tasks including robust learning such as noisy learning, adversarially robust learning; data-efficient learning such as semi-supervised and weakly-supervised learning; and model-efficient learning such as model compression and knowledge distillation.
Graph Neural Networks (GNNs) have attracted tremendous attention by demonstrating their capability to handle graph data. However, they are difficult to be deployed in resource-limited devices due to model sizes and scalability constraints imposed by the multi-hop data dependency. In addition, real-world graphs usually possess complex structural information and features. Therefore, to improve the applicability of GNNs and fully encode the complicated topological information, knowledge distillation on graphs (KDG) has been introduced to build a smaller yet effective model and exploit more knowledge from data, leading to model compression and performance improvement. Recently, KDG has achieved considerable progress with many studies proposed. In this survey, we systematically review these works. Specifically, we first introduce KDG challenges and bases, then categorize and summarize existing works of KDG by answering the following three questions: 1) what to distillate, 2) who to whom, and 3) how to distillate. Finally, we share our thoughts on future research directions.
Real-world data contains a vast amount of multimodal information, among which vision and language are the two most representative modalities. Moreover, increasingly heavier models, \textit{e}.\textit{g}., Transformers, have attracted the attention of researchers to model compression. However, how to compress multimodal models, especially vison-language Transformers, is still under-explored. This paper proposes the \textbf{U}nified and \textbf{P}r\textbf{o}gressive \textbf{P}runing (\textbf{\emph{UPop}}) as a universal vison-language Transformer compression framework, which incorporates 1) unifiedly searching multimodal subnets in a continuous optimization space from the original model, which enables automatic assignment of pruning ratios among compressible modalities and structures; 2) progressively searching and retraining the subnet, which maintains convergence between the search and retrain to attain higher compression ratios. Experiments on various tasks, datasets, and model architectures demonstrate the effectiveness and versatility of the proposed UPop framework. The code is available at https://github.com/sdc17/UPop.
As a general model compression paradigm, feature-based knowledge distillation allows the student model to learn expressive features from the teacher counterpart. In this paper, we mainly focus on designing an effective feature-distillation framework and propose a spatial-channel adaptive masked distillation (AMD) network for object detection. More specifically, in order to accurately reconstruct important feature regions, we first perform attention-guided feature masking on the feature map of the student network, such that we can identify the important features via spatially adaptive feature masking instead of random masking in the previous methods. In addition, we employ a simple and efficient module to allow the student network channel to be adaptive, improving its model capability in object perception and detection. In contrast to the previous methods, more crucial object-aware features can be reconstructed and learned from the proposed network, which is conducive to accurate object detection. The empirical experiments demonstrate the superiority of our method: with the help of our proposed distillation method, the student networks report 41.3%, 42.4%, and 42.7% mAP scores when RetinaNet, Cascade Mask-RCNN and RepPoints are respectively used as the teacher framework for object detection, which outperforms the previous state-of-the-art distillation methods including FGD and MGD.
Knowledge distillation (KD) is one of the prominent techniques for model compression. In this method, the knowledge of a large network (teacher) is distilled into a model (student) with usually significantly fewer parameters. KD tries to better-match the output of the student model to that of the teacher model based on the knowledge extracts from the forward pass of the teacher network. Although conventional KD is effective for matching the two networks over the given data points, there is no guarantee that these models would match in other areas for which we do not have enough training samples. In this work, we address that problem by generating new auxiliary training samples based on extracting knowledge from the backward pass of the teacher in the areas where the student diverges greatly from the teacher. We compute the difference between the teacher and the student and generate new data samples that maximize the divergence. This is done by perturbing data samples in the direction of the gradient of the difference between the student and the teacher. Augmenting the training set by adding this auxiliary improves the performance of KD significantly and leads to a closer match between the student and the teacher. Using this approach, when data samples come from a discrete domain, such as applications of natural language processing (NLP) and language understanding, is not trivial. However, we show how this technique can be used successfully in such applications. We evaluated the performance of our method on various tasks in computer vision and NLP domains and got promising results.
Low-rank compression is an important model compression strategy for obtaining compact neural network models. In general, because the rank values directly determine the model complexity and model accuracy, proper selection of layer-wise rank is very critical and desired. To date, though many low-rank compression approaches, either selecting the ranks in a manual or automatic way, have been proposed, they suffer from costly manual trials or unsatisfied compression performance. In addition, all of the existing works are not designed in a hardware-aware way, limiting the practical performance of the compressed models on real-world hardware platforms. To address these challenges, in this paper we propose HALOC, a hardware-aware automatic low-rank compression framework. By interpreting automatic rank selection from an architecture search perspective, we develop an end-to-end solution to determine the suitable layer-wise ranks in a differentiable and hardware-aware way. We further propose design principles and mitigation strategy to efficiently explore the rank space and reduce the potential interference problem. Experimental results on different datasets and hardware platforms demonstrate the effectiveness of our proposed approach. On CIFAR-10 dataset, HALOC enables 0.07% and 0.38% accuracy increase over the uncompressed ResNet-20 and VGG-16 models with 72.20% and 86.44% fewer FLOPs, respectively. On ImageNet dataset, HALOC achieves 0.9% higher top-1 accuracy than the original ResNet-18 model with 66.16% fewer FLOPs. HALOC also shows 0.66% higher top-1 accuracy increase than the state-of-the-art automatic low-rank compression solution with fewer computational and memory costs. In addition, HALOC demonstrates the practical speedups on different hardware platforms, verified by the measurement results on desktop GPU, embedded GPU and ASIC accelerator.
Ensemble learning has gain attention in resent deep learning research as a way to further boost the accuracy and generalizability of deep neural network (DNN) models. Recent ensemble training method explores different training algorithms or settings on multiple sub-models with the same model architecture, which lead to significant burden on memory and computation cost of the ensemble model. Meanwhile, the heurtsically induced diversity may not lead to significant performance gain. We propose a new prespective on exploring the intrinsic diversity within a model architecture to build efficient DNN ensemble. We make an intriguing observation that pruning and quantization, while both leading to efficient model architecture at the cost of small accuracy drop, leads to distinct behavior in the decision boundary. To this end, we propose Heterogeneously Compressed Ensemble (HCE), where we build an efficient ensemble with the pruned and quantized variants from a pretrained DNN model. An diversity-aware training objective is proposed to further boost the performance of the HCE ensemble. Experiemnt result shows that HCE achieves significant improvement in the efficiency-accuracy tradeoff comparing to both traditional DNN ensemble training methods and previous model compression methods.
Model compression via quantization and sparsity enhancement has gained an immense interest to enable the deployment of deep neural networks (DNNs) in resource-constrained edge environments. Although these techniques have shown promising results in reducing the energy, latency and memory requirements of the DNNs, their performance in non-ideal real-world settings (such as in the presence of hardware faults) is yet to be completely understood. In this paper, we investigate the impact of bit-flip and stuck-at faults on activation-sparse quantized DNNs (QDNNs). We show that a high level of activation sparsity comes at the cost of larger vulnerability to faults. For instance, activation-sparse QDNNs exhibit up to 17.32% lower accuracy than the standard QDNNs. We also establish that one of the major cause of the degraded accuracy is sharper minima in the loss landscape for activation-sparse QDNNs, which makes them more sensitive to perturbations in the weight values due to faults. Based on this observation, we propose the mitigation of the impact of faults by employing a sharpness-aware quantization (SAQ) training scheme. The activation-sparse and standard QDNNs trained with SAQ have up to 36.71% and 24.76% higher inference accuracy, respectively compared to their conventionally trained equivalents. Moreover, we show that SAQ-trained activation-sparse QDNNs show better accuracy in faulty settings than standard QDNNs trained conventionally. Thus the proposed technique can be instrumental in achieving sparsity-related energy/latency benefits without compromising on fault tolerance.