While knowledge distillation (transfer) has been attracting attentions from the research community, the recent development in the fields has heightened the need for reproducible studies and highly generalized frameworks to lower barriers to such high-quality, reproducible deep learning research. Several researchers voluntarily published frameworks used in their knowledge distillation studies to help other interested researchers reproduce their original work. Such frameworks, however, are usually neither well generalized nor maintained, thus researchers are still required to write a lot of code to refactor/build on the frameworks for introducing new methods, models, datasets and designing experiments. In this paper, we present our developed open-source framework built on PyTorch and dedicated for knowledge distillation studies. The framework is designed to enable users to design experiments by declarative PyYAML configuration files, and helps researchers complete the recently proposed ML Code Completeness Checklist. Using the developed framework, we demonstrate its various efficient training strategies, and implement a variety of knowledge distillation methods. We also reproduce some of their original experimental results on the ImageNet and COCO datasets presented at major machine learning conferences such as ICLR, NeurIPS, CVPR and ECCV, including recent state-of-the-art methods. All the source code, configurations, log files and trained model weights are publicly available at https://github.com/yoshitomo-matsubara/torchdistill .
Model compression aims to deploy deep neural networks (DNN) on mobile devices with limited computing and storage resources. However, most of the existing model compression methods rely on manually defined rules, which require domain expertise. DNNs are essentially computational graphs, which contain rich structural information. In this paper, we aim to find a suitable compression policy from DNNs' structural information. We propose an automatic graph encoder-decoder model compression (AGMC) method combined with graph neural networks (GNN) and reinforcement learning (RL). We model the target DNN as a graph and use GNN to learn the DNN's embeddings automatically. We compared our method with rule-based DNN embedding model compression methods to show the effectiveness of our method. Results show that our learning-based DNN embedding achieves better performance and a higher compression ratio with fewer search steps. We evaluated our method on over-parameterized and mobile-friendly DNNs and compared our method with handcrafted and learning-based model compression approaches. On over parameterized DNNs, such as ResNet-56, our method outperformed handcrafted and learning-based methods with $4.36\%$ and $2.56\%$ higher accuracy, respectively. Furthermore, on MobileNet-v2, we achieved a higher compression ratio than state-of-the-art methods with just $0.93\%$ accuracy loss.
Current low-precision quantization algorithms often have the hidden cost of conversion back and forth from floating point to quantized integer values. This hidden cost limits the latency improvement realized by quantizing Neural Networks. To address this, we present HAWQV3, a novel mixed-precision integer-only quantization framework. The contributions of HAWQV3 are the following: (i) An integer-only inference where the entire computational graph is performed only with integer multiplication, addition, and bit shifting, without any floating point operations or even integer division; (ii) A novel hardware-aware mixed-precision quantization method where the bit-precision is calculated by solving an integer linear programming problem that balances the trade-off between model perturbation and other constraints, e.g., memory footprint and latency; (iii) Direct hardware deployment and open source contribution for 4-bit uniform/mixed-precision quantization in TVM, achieving an average speed up of $1.45\times$ for uniform 4-bit, as compared to uniform 8-bit for ResNet50 on T4 GPUs; and (iv) extensive evaluation of the proposed methods on ResNet18/50 and InceptionV3, for various model compression levels with/without mixed precision. For ResNet50, our INT8 quantization achieves an accuracy of $77.58\%$, which is $2.68\%$ higher than prior integer-only work, and our mixed-precision INT4/8 quantization can reduce INT8 latency by $23\%$ and still achieve $76.73\%$ accuracy. Our framework and the TVM implementation have been open sourced.
WaveNet is a state-of-the-art text-to-speech vocoder that remains challenging to deploy due to its autoregressive loop. In this work we focus on ways to accelerate the original WaveNet architecture directly, as opposed to modifying the architecture, such that the model can be deployed as part of a scalable text-to-speech system. We survey a wide variety of model compression techniques that are amenable to deployment on a range of hardware platforms. In particular, we compare different model sparsity methods and levels, and seven widely used precisions as targets for quantization; and are able to achieve models with a compression ratio of up to 13.84 without loss in audio fidelity compared to a dense, single-precision floating-point baseline. All techniques are implemented using existing open source deep learning frameworks and libraries to encourage their wider adoption.
User data confidentiality protection is becoming a rising challenge in the present deep learning research. Without access to data, conventional data-driven model compression faces a higher risk of performance degradation. Recently, some works propose to generate images from a specific pretrained model to serve as training data. However, the inversion process only utilizes biased feature statistics stored in one model and is from low-dimension to high-dimension. As a consequence, it inevitably encounters the difficulties of generalizability and inexact inversion, which leads to unsatisfactory performance. To address these problems, we propose MixMix based on two simple yet effective techniques: (1) Feature Mixing: utilizes various models to construct a universal feature space for generalized inversion; (2) Data Mixing: mixes the synthesized images and labels to generate exact label information. We prove the effectiveness of MixMix from both theoretical and empirical perspectives. Extensive experiments show that MixMix outperforms existing methods on the mainstream compression tasks, including quantization, knowledge distillation, and pruning. Specifically, MixMix achieves up to 4% and 20% accuracy uplift on quantization and pruning, respectively, compared to existing data-free compression work.
This paper presents a novel knowledge distillation based model compression framework consisting of a student ensemble. It enables distillation of simultaneously learnt ensemble knowledge onto each of the compressed student models. Each model learns unique representations from the data distribution due to its distinct architecture. This helps the ensemble generalize better by combining every model's knowledge. The distilled students and ensemble teacher are trained simultaneously without requiring any pretrained weights. Moreover, our proposed method can deliver multi-compressed students with single training, which is efficient and flexible for different scenarios. We provide comprehensive experiments using state-of-the-art classification models to validate our framework's effectiveness. Notably, using our framework a 97% compressed ResNet110 student model managed to produce a 10.64% relative accuracy gain over its individual baseline training on CIFAR100 dataset. Similarly a 95% compressed DenseNet-BC(k=12) model managed a 8.17% relative accuracy gain.
In the traditional deep compression framework, iteratively performing network pruning and quantization can reduce the model size and computation cost to meet the deployment requirements. However, such a step-wise application of pruning and quantization may lead to suboptimal solutions and unnecessary time consumption. In this paper, we tackle this issue by integrating network pruning and quantization as a unified joint compression problem and then use AutoML to automatically solve it. We find the pruning process can be regarded as the channel-wise quantization with 0 bit. Thus, the separate two-step pruning and quantization can be simplified as the one-step quantization with mixed precision. This unification not only simplifies the compression pipeline but also avoids the compression divergence. To implement this idea, we propose the automated model compression by jointly applied pruning and quantization (AJPQ). AJPQ is designed with a hierarchical architecture: the layer controller controls the layer sparsity, and the channel controller decides the bit-width for each kernel. Following the same importance criterion, the layer controller and the channel controller collaboratively decide the compression strategy. With the help of reinforcement learning, our one-step compression is automatically achieved. Compared with the state-of-the-art automated compression methods, our method obtains a better accuracy while reducing the storage considerably. For fixed precision quantization, AJPQ can reduce more than five times model size and two times computation with a slight performance increase for Skynet in remote sensing object detection. When mixed-precision is allowed, AJPQ can reduce five times model size with only 1.06% top-5 accuracy decline for MobileNet in the classification task.
Previous studies on image classification have mainly focused on the performance of the networks, not on real-time operation or model compression. We propose a Gaussian Deep Recurrent visual Attention Model (GDRAM)- a reinforcement learning based lightweight deep neural network for large scale image classification that outperforms the conventional CNN (Convolutional Neural Network) which uses the entire image as input. Highly inspired by the biological visual recognition process, our model mimics the stochastic location of the retina with Gaussian distribution. We evaluate the model on Large cluttered MNIST, Large CIFAR-10 and Large CIFAR-100 datasets which are resized to 128 in both width and height.
Automated Machine Learning(Auto-ML) pruning methods aim at searching a pruning strategy automatically to reduce the computational complexity of deep Convolutional Neural Networks(deep CNNs). However, some previous work found that the results of many Auto-ML pruning methods cannot even surpass the results of the uniformly pruning method. In this paper, the ineffectiveness of Auto-ML pruning which is caused by unfull and unfair training of the supernet is shown. A deep supernet suffers from unfull training because it contains too many candidates. To overcome the unfull training, a stage-wise pruning(SWP) method is proposed, which splits a deep supernet into several stage-wise supernets to reduce the candidate number and utilize inplace distillation to supervise the stage training. Besides, A wide supernet is hit by unfair training since the sampling probability of each channel is unequal. Therefore, the fullnet and the tinynet are sampled in each training iteration to ensure each channel can be overtrained. Remarkably, the proxy performance of the subnets trained with SWP is closer to the actual performance than that of most of the previous Auto-ML pruning work. Experiments show that SWP achieves the state-of-the-art on both CIFAR-10 and ImageNet under the mobile setting.
The compression of deep neural networks (DNNs) to reduce inference cost becomes increasingly important to meet realistic deployment requirements of various applications. There have been a significant amount of work regarding network compression, while most of them are heuristic rule-based or typically not friendly to be incorporated into varying scenarios. On the other hand, sparse optimization yielding sparse solutions naturally fits the compression requirement, but due to the limited study of sparse optimization in stochastic learning, its extension and application onto model compression is rarely well explored. In this work, we propose a model compression framework based on the recent progress on sparse stochastic optimization. Compared to existing model compression techniques, our method is effective and requires fewer extra engineering efforts to incorporate with varying applications, and has been numerically demonstrated on benchmark compression tasks. Particularly, we achieve up to 7.2 and 2.9 times FLOPs reduction with the same level of evaluation accuracy on VGG16 for CIFAR10 and ResNet50 for ImageNet compared to the baseline heavy models, respectively.
Deep pre-trained Transformer models have achieved state-of-the-art results over a variety of natural language processing (NLP) tasks. By learning rich language knowledge with millions of parameters, these models are usually overparameterized and significantly increase the computational overhead in applications. It is intuitive to address this issue by model compression. In this work, we propose a method, called Single-Shot Meta-Pruning, to compress deep pre-trained Transformers before fine-tuning. Specifically, we focus on pruning unnecessary attention heads adaptively for different downstream tasks. To measure the informativeness of attention heads, we train our Single-Shot Meta-Pruner (SMP) with a meta-learning paradigm aiming to maintain the distribution of text representations after pruning. Compared with existing compression methods for pre-trained models, our method can reduce the overhead of both fine-tuning and inference. Experimental results show that our pruner can selectively prune 50% of attention heads with little impact on the performance on downstream tasks and even provide better text representations. The source code will be released in the future.
Knowledge distillation (KD) has enabled remarkable progress in model compression and knowledge transfer. However, KD requires a large volume of original data or their representation statistics that are not usually available in practice. Data-free KD has recently been proposed to resolve this problem, wherein teacher and student models are fed by a synthetic sample generator trained from the teacher. Nonetheless, existing data-free KD methods rely on fine-tuning of weights to balance multiple losses, and ignore the diversity of generated samples, resulting in limited accuracy and robustness. To overcome this challenge, we propose robustness and diversity seeking data-free KD (RDSKD) in this paper. The generator loss function is crafted to produce samples with high authenticity, class diversity, and inter-sample diversity. Without real data, the objectives of seeking high sample authenticity and class diversity often conflict with each other, causing frequent loss fluctuations. We mitigate this by exponentially penalizing loss increments. With MNIST, CIFAR-10, and SVHN datasets, our experiments show that RDSKD achieves higher accuracy with more robustness over different hyperparameter settings, compared to other data-free KD methods such as DAFL, MSKD, ZSKD, and DeepInversion.
Singing Voice Detection (SVD) has been an active area of research in music information retrieval (MIR). Currently, two deep neural network-based methods, one based on CNN and the other on RNN, exist in literature that learn optimized features for the voice detection (VD) task and achieve state-of-the-art performance on common datasets. Both these models have a huge number of parameters (1.4M for CNN and 65.7K for RNN) and hence not suitable for deployment on devices like smartphones or embedded sensors with limited capacity in terms of memory and computation power. The most popular method to address this issue is known as knowledge distillation in deep learning literature (in addition to model compression) where a large pre-trained network known as the teacher is used to train a smaller student network. Given the wide applications of SVD in music information retrieval, to the best of our knowledge, model compression for practical deployment has not yet been explored. In this paper, efforts have been made to investigate this issue using both conventional as well as ensemble knowledge distillation techniques.
This paper presents a dynamic network rewiring (DNR) method to generate pruned deep neural network (DNN) models that are robust against adversarial attacks yet maintain high accuracy on clean images. In particular, the disclosed DNR method is based on a unified constrained optimization formulation using a hybrid loss function that merges ultra-high model compression with robust adversarial training. This training strategy dynamically adjusts inter-layer connectivity based on per-layer normalized momentum computed from the hybrid loss function. In contrast to existing robust pruning frameworks that require multiple training iterations, the proposed learning strategy achieves an overall target pruning ratio with only a single training iteration and can be tuned to support both irregular and structured channel pruning. To evaluate the merits of DNR, experiments were performed with two widely accepted models, namely VGG16 and ResNet-18, on CIFAR-10, CIFAR-100 as well as with VGG16 on Tiny-ImageNet. Compared to the baseline uncompressed models, DNR provides over20x compression on all the datasets with no significant drop in either clean or adversarial classification accuracy. Moreover, our experiments show that DNR consistently finds compressed models with better clean and adversarial image classification performance than what is achievable through state-of-the-art alternatives.
This work explores how self-supervised learning can be universally used to discover speaker-specific features towards enabling personalized speech enhancement models. We specifically address the few-shot learning scenario where access to cleaning recordings of a test-time speaker is limited to a few seconds, but noisy recordings of the speaker are abundant. We develop a simple contrastive learning procedure which treats the abundant noisy data as makeshift training targets through pairwise noise injection: the model is pretrained to maximize agreement between pairs of differently deformed identical utterances and to minimize agreement between pairs of similarly deformed nonidentical utterances. Our experiments compare the proposed pretraining approach with two baseline alternatives: speaker-agnostic fully-supervised pretraining, and speaker-specific self-supervised pretraining without contrastive loss terms. Of all three approaches, the proposed method using contrastive mixtures is found to be most robust to model compression (using 85% fewer parameters) and reduced clean speech (requiring only 3 seconds).
Automated Machine Learning (AutoML) is an important industrial solution for automatic discovery and deployment of the machine learning models. However, designing an integrated AutoML system faces four great challenges of configurability, scalability, integrability, and platform diversity. In this work, we present VEGA, an efficient and comprehensive AutoML framework that is compatible and optimized for multiple hardware platforms. a) The VEGA pipeline integrates various modules of AutoML, including Neural Architecture Search (NAS), Hyperparameter Optimization (HPO), Auto Data Augmentation, Model Compression, and Fully Train. b) To support a variety of search algorithms and tasks, we design a novel fine-grained search space and its description language to enable easy adaptation to different search algorithms and tasks. c) We abstract the common components of deep learning frameworks into a unified interface. VEGA can be executed with multiple back-ends and hardwares. Extensive benchmark experiments on multiple tasks demonstrate that VEGA can improve the existing AutoML algorithms and discover new high-performance models against SOTA methods, e.g. the searched DNet model zoo for Ascend 10x faster than EfficientNet-B5 and 9.2x faster than RegNetX-32GF on ImageNet. VEGA is open-sourced at https://github.com/huawei-noah/vega.
Many learning tasks require us to deal with graph data which contains rich relational information among elements, leading increasing graph neural network (GNN) models to be deployed in industrial products for improving the quality of service. However, they also raise challenges to model authentication. It is necessary to protect the ownership of the GNN models, which motivates us to present a watermarking method to GNN models in this paper. In the proposed method, an Erdos-Renyi (ER) random graph with random node feature vectors and labels is randomly generated as a trigger to train the GNN to be protected together with the normal samples. During model training, the secret watermark is embedded into the label predictions of the ER graph nodes. During model verification, by activating a marked GNN with the trigger ER graph, the watermark can be reconstructed from the output to verify the ownership. Since the ER graph was randomly generated, by feeding it to a non-marked GNN, the label predictions of the graph nodes are random, resulting in a low false alarm rate (of the proposed work). Experimental results have also shown that, the performance of a marked GNN on its original task will not be impaired. Moreover, it is robust against model compression and fine-tuning, which has shown the superiority and applicability.
Despite tremendous success in many application scenarios, deep learning faces serious intellectual property (IP) infringement threats. Considering the cost of designing and training a good model, infringements will significantly infringe the interests of the original model owner. Recently, many impressive works have emerged for deep model IP protection. However, they either are vulnerable to ambiguity attacks, or require changes in the target network structure by replacing its original normalization layers and hence cause significant performance drops. To this end, we propose a new passport-aware normalization formulation, which is generally applicable to most existing normalization layers and only needs to add another passport-aware branch for IP protection. This new branch is jointly trained with the target model but discarded in the inference stage. Therefore it causes no structure change in the target model. Only when the model IP is suspected to be stolen by someone, the private passport-aware branch is added back for ownership verification. Through extensive experiments, we verify its effectiveness in both image and 3D point recognition models. It is demonstrated to be robust not only to common attack techniques like fine-tuning and model compression, but also to ambiguity attacks. By further combining it with trigger-set based methods, both black-box and white-box verification can be achieved for enhanced security of deep learning models deployed in real systems. Code can be found at https://github.com/ZJZAC/Passport-aware-Normalization.
Self-supervised learning aims to learn good representations with unlabeled data. Recent works have shown that larger models benefit more from self-supervised learning than smaller models. As a result, the gap between supervised and self-supervised learning has been greatly reduced for larger models. In this work, instead of designing a new pseudo task for self-supervised learning, we develop a model compression method to compress an already learned, deep self-supervised model (teacher) to a smaller one (student). We train the student model so that it mimics the relative similarity between the data points in the teacher's embedding space. For AlexNet, our method outperforms all previous methods including the fully supervised model on ImageNet linear evaluation (59.0% compared to 56.5%) and on nearest neighbor evaluation (50.7% compared to 41.4%). To the best of our knowledge, this is the first time a self-supervised AlexNet has outperformed supervised one on ImageNet classification. Our code is available here: https://github.com/UMBCvision/CompRess
Convolutional neural networks (CNNs) play a key role in deep learning applications. However, the large storage overheads and the substantial computation cost of CNNs are problematic in hardware accelerators. Computing-in-memory (CIM) architecture has demonstrated great potential to effectively compute large-scale matrix-vector multiplication. However, the intensive multiply and accumulation (MAC) operations executed at the crossbar array and the limited capacity of CIM macros remain bottlenecks for further improvement of energy efficiency and throughput. To reduce computation costs, network pruning and quantization are two widely studied compression methods to shrink the model size. However, most of the model compression algorithms can only be implemented in digital-based CNN accelerators. For implementation in a static random access memory (SRAM) CIM-based accelerator, the model compression algorithm must consider the hardware limitations of CIM macros, such as the number of word lines and bit lines that can be turned on at the same time, as well as how to map the weight to the SRAM CIM macro. In this study, a software and hardware co-design approach is proposed to design an SRAM CIM-based CNN accelerator and an SRAM CIM-aware model compression algorithm. To lessen the high-precision MAC required by batch normalization (BN), a quantization algorithm that can fuse BN into the weights is proposed. Furthermore, to reduce the number of network parameters, a sparsity algorithm that considers a CIM architecture is proposed. Last, MARS, a CIM-based CNN accelerator that can utilize multiple SRAM CIM macros as processing units and support a sparsity neural network, is proposed.