Deep neural networks have exhibited remarkable performance in image super-resolution (SR) tasks by learning a mapping from low-resolution (LR) images to high-resolution (HR) images. However, the SR problem is typically an ill-posed problem and existing methods would come with several limitations. First, the possible mapping space of SR can be extremely large since there may exist many different HR images that can be super-resolved from the same LR image. As a result, it is hard to directly learn a promising SR mapping from such a large space. Second, it is often inevitable to develop very large models with extremely high computational cost to yield promising SR performance. In practice, one can use model compression techniques to obtain compact models by reducing model redundancy. Nevertheless, it is hard for existing model compression methods to accurately identify the redundant components due to the extremely large SR mapping space. To alleviate the first challenge, we propose a dual regression learning scheme to reduce the space of possible SR mappings. Specifically, in addition to the mapping from LR to HR images, we learn an additional dual regression mapping to estimate the downsampling kernel and reconstruct LR images. In this way, the dual mapping acts as a constraint to reduce the space of possible mappings. To address the second challenge, we propose a dual regression compression (DRC) method to reduce model redundancy in both layer-level and channel-level based on channel pruning. Specifically, we first develop a channel number search method that minimizes the dual regression loss to determine the redundancy of each layer. Given the searched channel numbers, we further exploit the dual regression manner to evaluate the importance of channels and prune the redundant ones. Extensive experiments show the effectiveness of our method in obtaining accurate and efficient SR models.
In 6G mobile communication systems, various AI-based network functions and applications have been standardized. Federated learning (FL) is adopted as the core learning architecture for 6G systems to avoid privacy leakage from mobile user data. However, in FL, users with non-independent and identically distributed (non-IID) datasets can deteriorate the performance of the global model because the convergence direction of the gradient for each dataset is different, thereby inducing a weight divergence problem. To address this problem, we propose a novel diffusion strategy for machine learning (ML) models (FedDif) to maximize the performance of the global model with non-IID data. FedDif enables the local model to learn different distributions before parameter aggregation by passing the local models through users via device-to-device communication. Furthermore, we theoretically demonstrate that FedDif can circumvent the weight-divergence problem. Based on this theory, we propose a communication-efficient diffusion strategy for ML models that can determine the trade-off between learning performance and communication cost using auction theory. The experimental results show that FedDif improves the top-1 test accuracy by up to 34.89\% and reduces communication costs by 14.6% to a maximum of 63.49%.
We tackle the essential task of finding dense visual correspondences between a pair of images. This is a challenging problem due to various factors such as poor texture, repetitive patterns, illumination variation, and motion blur in practical scenarios. In contrast to methods that use dense correspondence ground-truths as direct supervision for local feature matching training, we train 3DG-STFM: a multi-modal matching model (Teacher) to enforce the depth consistency under 3D dense correspondence supervision and transfer the knowledge to 2D unimodal matching model (Student). Both teacher and student models consist of two transformer-based matching modules that obtain dense correspondences in a coarse-to-fine manner. The teacher model guides the student model to learn RGB-induced depth information for the matching purpose on both coarse and fine branches. We also evaluate 3DG-STFM on a model compression task. To the best of our knowledge, 3DG-STFM is the first student-teacher learning method for the local feature matching task. The experiments show that our method outperforms state-of-the-art methods on indoor and outdoor camera pose estimations, and homography estimation problems. Code is available at: https://github.com/Ryan-prime/3DG-STFM.
Mobile devices run deep learning models for various purposes, such as image classification and speech recognition. Due to the resource constraints of mobile devices, researchers have focused on either making a lightweight deep neural network (DNN) model using model pruning or generating an efficient code using compiler optimization. Surprisingly, we found that the straightforward integration between model compression and compiler auto-tuning often does not produce the most efficient model for a target device. We propose CPrune, a compiler-informed model pruning for efficient target-aware DNN execution to support an application with a required target accuracy. CPrune makes a lightweight DNN model through informed pruning based on the structural information of subgraphs built during the compiler tuning process. Our experimental results show that CPrune increases the DNN execution speed up to 2.73x compared to the state-of-the-art TVM auto-tune while satisfying the accuracy requirement.
Recent studies shows that the majority of existing deep steganalysis models have a large amount of redundancy, which leads to a huge waste of storage and computing resources. The existing model compression method cannot flexibly compress the convolutional layer in residual shortcut block so that a satisfactory shrinking rate cannot be obtained. In this paper, we propose STD-NET, an unsupervised deep-learning architecture search approach via hierarchical tensor decomposition for image steganalysis. Our proposed strategy will not be restricted by various residual connections, since this strategy does not change the number of input and output channels of the convolution block. We propose a normalized distortion threshold to evaluate the sensitivity of each involved convolutional layer of the base model to guide STD-NET to compress target network in an efficient and unsupervised approach, and obtain two network structures of different shapes with low computation cost and similar performance compared with the original one. Extensive experiments have confirmed that, on one hand, our model can achieve comparable or even better detection performance in various steganalytic scenarios due to the great adaptivity of the obtained network architecture. On the other hand, the experimental results also demonstrate that our proposed strategy is more efficient and can remove more redundancy compared with previous steganalytic network compression methods.
Despite the popularity of Model Compression and Multitask Learning, how to effectively compress a multitask model has been less thoroughly analyzed due to the challenging entanglement of tasks in the parameter space. In this paper, we propose DiSparse, a simple, effective, and first-of-its-kind multitask pruning and sparse training scheme. We consider each task independently by disentangling the importance measurement and take the unanimous decisions among all tasks when performing parameter pruning and selection. Our experimental results demonstrate superior performance on various configurations and settings compared to popular sparse training and pruning methods. Besides the effectiveness in compression, DiSparse also provides a powerful tool to the multitask learning community. Surprisingly, we even observed better performance than some dedicated multitask learning methods in several cases despite the high model sparsity enforced by DiSparse. We analyzed the pruning masks generated with DiSparse and observed strikingly similar sparse network architecture identified by each task even before the training starts. We also observe the existence of a "watershed" layer where the task relatedness sharply drops, implying no benefits in continued parameters sharing. Our code and models will be available at: https://github.com/SHI-Labs/DiSparse-Multitask-Model-Compression.
We introduce canonical weight normalization for convolutional neural networks. Inspired by the canonical tensor decomposition, we express the weight tensors in so-called canonical networks as scaled sums of outer vector products. In particular, we train network weights in the decomposed form, where scale weights are optimized separately for each mode. Additionally, similarly to weight normalization, we include a global scaling parameter. We study the initialization of the canonical form by running the power method and by drawing randomly from Gaussian or uniform distributions. Our results indicate that we can replace the power method with cheaper initializations drawn from standard distributions. The canonical re-parametrization leads to competitive normalization performance on the MNIST, CIFAR10, and SVHN data sets. Moreover, the formulation simplifies network compression. Once training has converged, the canonical form allows convenient model-compression by truncating the parameter sums.
Recent studies have uncovered that language model distillation is less effective when facing a large capacity gap between the teacher and the student, and introduced teacher assistant-based distillation to bridge the gap. As a connection, the scale and the performance of the teacher assistant is of vital importance to bring the knowledge from the teacher to the student. However, existing teacher assistant-based methods require maximally many trials before scheduling an optimal teacher assistant. To this end, we propose a minimal distillation schedule (MiniDisc) for scheduling the optimal teacher assistant in minimally one trial. In particular, motivated by the finding that the performance of the student is positively correlated to the scale-performance tradeoff of the teacher assistant, MiniDisc is designed with a $\lambda$-tradeoff to measure the optimality of the teacher assistant without trial distillation to the student. MiniDisc then can schedule the optimal teacher assistant with the best $\lambda$-tradeoff in a sandwich framework. MiniDisc is evaluated with an extensive set of experiments on GLUE. Experimental results demonstrate the improved efficiency our MiniDisc compared to several state-of-the-art baselines. We further apply MiniDisc to a language model with billions of parameters and show its scalability.
Training deep reinforcement learning (DRL) models usually requires high computation costs. Therefore, compressing DRL models possesses immense potential for training acceleration and model deployment. However, existing methods that generate small models mainly adopt the knowledge distillation-based approach by iteratively training a dense network. As a result, the training process still demands massive computing resources. Indeed, sparse training from scratch in DRL has not been well explored and is particularly challenging due to non-stationarity in bootstrap training. In this work, we propose a novel sparse DRL training framework, "the Rigged Reinforcement Learning Lottery" (RLx2), which builds upon gradient-based topology evolution and is capable of training a sparse DRL model based entirely on a sparse network. Specifically, RLx2 introduces a novel multi-step TD target mechanism with a dynamic-capacity replay buffer to achieve robust value learning and efficient topology exploration in sparse models. It also reaches state-of-the-art sparse training performance in several tasks, showing 7.5\times-20\times model compression with less than 3% performance degradation and up to 20\times and 50\times FLOPs reduction for training and inference, respectively.
Despite substantial progress in 3D object detection, advanced 3D detectors often suffer from heavy computation overheads. To this end, we explore the potential of knowledge distillation (KD) for developing efficient 3D object detectors, focusing on popular pillar- and voxel-based detectors.In the absence of well-developed teacher-student pairs, we first study how to obtain student models with good trade offs between accuracy and efficiency from the perspectives of model compression and input resolution reduction. Then, we build a benchmark to assess existing KD methods developed in the 2D domain for 3D object detection upon six well-constructed teacher-student pairs. Further, we propose an improved KD pipeline incorporating an enhanced logit KD method that performs KD on only a few pivotal positions determined by teacher classification response, and a teacher-guided student model initialization to facilitate transferring teacher model's feature extraction ability to students through weight inheritance. Finally, we conduct extensive experiments on the Waymo dataset. Our best performing model achieves $65.75\%$ LEVEL 2 mAPH, surpassing its teacher model and requiring only $44\%$ of teacher flops. Our most efficient model runs 51 FPS on an NVIDIA A100, which is $2.2\times$ faster than PointPillar with even higher accuracy. Code is available at \url{https://github.com/CVMI-Lab/SparseKD}.
The remarkable breakthroughs in point cloud representation learning have boosted their usage in real-world applications such as self-driving cars and virtual reality. However, these applications usually have an urgent requirement for not only accurate but also efficient 3D object detection. Recently, knowledge distillation has been proposed as an effective model compression technique, which transfers the knowledge from an over-parameterized teacher to a lightweight student and achieves consistent effectiveness in 2D vision. However, due to point clouds' sparsity and irregularity, directly applying previous image-based knowledge distillation methods to point cloud detectors usually leads to unsatisfactory performance. To fill the gap, this paper proposes PointDistiller, a structured knowledge distillation framework for point clouds-based 3D detection. Concretely, PointDistiller includes local distillation which extracts and distills the local geometric structure of point clouds with dynamic graph convolution and reweighted learning strategy, which highlights student learning on the crucial points or voxels to improve knowledge distillation efficiency. Extensive experiments on both voxels-based and raw points-based detectors have demonstrated the effectiveness of our method over seven previous knowledge distillation methods. For instance, our 4X compressed PointPillars student achieves 2.8 and 3.4 mAP improvements on BEV and 3D object detection, outperforming its teacher by 0.9 and 1.8 mAP, respectively. Codes have been released at https://github.com/RunpeiDong/PointDistiller.
Black-Box Knowledge Distillation (B2KD) is a formulated problem for cloud-to-edge model compression with invisible data and models hosted on the server. B2KD faces challenges such as limited Internet exchange and edge-cloud disparity of data distributions. In this paper, we formalize a two-step workflow consisting of deprivatization and distillation, and theoretically provide a new optimization direction from logits to cell boundary different from direct logits alignment. With its guidance, we propose a new method Mapping-Emulation KD (MEKD) that distills a black-box cumbersome model into a lightweight one. Our method does not differentiate between treating soft or hard responses, and consists of: 1) deprivatization: emulating the inverse mapping of the teacher function with a generator, and 2) distillation: aligning low-dimensional logits of the teacher and student models by reducing the distance of high-dimensional image points. For different teacher-student pairs, our method yields inspiring distillation performance on various benchmarks, and outperforms the previous state-of-the-art approaches.
In this paper, we introduce InDistill, a method that serves as a warmup stage for enhancing Knowledge Distillation (KD) effectiveness. InDistill focuses on transferring critical information flow paths from a heavyweight teacher to a lightweight student. This is achieved via a training scheme based on curriculum learning that considers the distillation difficulty of each layer and the critical learning periods when the information flow paths are established. This procedure can lead to a student model that is better prepared to learn from the teacher. To ensure the applicability of InDistill across a wide range of teacher-student pairs, we also incorporate a pruning operation when there is a discrepancy in the width of the teacher and student layers. This pruning operation reduces the width of the teacher's intermediate layers to match those of the student, allowing direct distillation without the need for an encoding stage. The proposed method is extensively evaluated using various pairs of teacher-student architectures on CIFAR-10, CIFAR-100, and ImageNet datasets demonstrating that preserving the information flow paths consistently increases the performance of the baseline KD approaches on both classification and retrieval settings. The code is available at https://github.com/gsarridis/InDistill.
Transformer models have been developed in molecular science with excellent performance in applications including quantitative structure-activity relationship (QSAR) and virtual screening (VS). Compared with other types of models, however, they are large, which results in a high hardware requirement to abridge time for both training and inference processes. In this work, cross-layer parameter sharing (CLPS), and knowledge distillation (KD) are used to reduce the sizes of transformers in molecular science. Both methods not only have competitive QSAR predictive performance as compared to the original BERT model, but also are more parameter efficient. Furthermore, by integrating CLPS and KD into a two-state chemical network, we introduce a new deep lite chemical transformer model, DeLiCaTe. DeLiCaTe captures general-domains as well as task-specific knowledge, which lead to a 4x faster rate of both training and inference due to a 10- and 3-times reduction of the number of parameters and layers, respectively. Meanwhile, it achieves comparable performance in QSAR and VS modeling. Moreover, we anticipate that the model compression strategy provides a pathway to the creation of effective generative transformer models for organic drug and material design.
Current text-image approaches (e.g., CLIP) typically adopt dual-encoder architecture us- ing pre-trained vision-language representation. However, these models still pose non-trivial memory requirements and substantial incre- mental indexing time, which makes them less practical on mobile devices. In this paper, we present an effective two-stage framework to compress large pre-trained dual-encoder for lightweight text-image retrieval. The result- ing model is smaller (39% of the original), faster (1.6x/2.9x for processing image/text re- spectively), yet performs on par with or bet- ter than the original full model on Flickr30K and MSCOCO benchmarks. We also open- source an accompanying realistic mobile im- age search application.
Deep Neural Networks (DNNs) have gained considerable attention in the past decades due to their astounding performance in different applications, such as natural language modeling, self-driving assistance, and source code understanding. With rapid exploration, more and more complex DNN architectures have been proposed along with huge pre-trained model parameters. The common way to use such DNN models in user-friendly devices (e.g., mobile phones) is to perform model compression before deployment. However, recent research has demonstrated that model compression, e.g., model quantization, yields accuracy degradation as well as outputs disagreements when tested on unseen data. Since the unseen data always include distribution shifts and often appear in the wild, the quality and reliability of quantized models are not ensured. In this paper, we conduct a comprehensive study to characterize and help users understand the behaviors of quantized models. Our study considers 4 datasets spanning from image to text, 8 DNN architectures including feed-forward neural networks and recurrent neural networks, and 42 shifted sets with both synthetic and natural distribution shifts. The results reveal that 1) data with distribution shifts happen more disagreements than without. 2) Quantization-aware training can produce more stable models than standard, adversarial, and Mixup training. 3) Disagreements often have closer top-1 and top-2 output probabilities, and $Margin$ is a better indicator than the other uncertainty metrics to distinguish disagreements. 4) Retraining with disagreements has limited efficiency in removing disagreements. We opensource our code and models as a new benchmark for further studying the quantized models.
We introduce LilNetX, an end-to-end trainable technique for neural networks that enables learning models with specified accuracy-rate-computation trade-off. Prior works approach these problems one at a time and often require post-processing or multistage training which become less practical and do not scale very well for large datasets or architectures. Our method constructs a joint training objective that penalizes the self-information of network parameters in a reparameterized latent space to encourage small model size while also introducing priors to increase structured sparsity in the parameter space to reduce computation. We achieve up to 50% smaller model size and 98% model sparsity on ResNet-20 while retaining the same accuracy on the CIFAR-10 dataset as well as 35% smaller model size and 42% structured sparsity on ResNet-50 trained on ImageNet, when compared to existing state-of-the-art model compression methods. Code is available at https://github.com/Sharath-girish/LilNetX.
Adversarial examples provoke weak reliability and potential security issues in deep neural networks. Although adversarial training has been widely studied to improve adversarial robustness, it works in an over-parameterized regime and requires high computations and large memory budgets. To bridge adversarial robustness and model compression, we propose a novel adversarial pruning method, Masking Adversarial Damage (MAD) that employs second-order information of adversarial loss. By using it, we can accurately estimate adversarial saliency for model parameters and determine which parameters can be pruned without weakening adversarial robustness. Furthermore, we reveal that model parameters of initial layer are highly sensitive to the adversarial examples and show that compressed feature representation retains semantic information for the target objects. Through extensive experiments on three public datasets, we demonstrate that MAD effectively prunes adversarially trained networks without loosing adversarial robustness and shows better performance than previous adversarial pruning methods.
The growing size of neural language models has led to increased attention in model compression. The two predominant approaches are pruning, which gradually removes weights from a pre-trained model, and distillation, which trains a smaller compact model to match a larger one. Pruning methods can significantly reduce the model size but hardly achieve large speedups as distillation. However, distillation methods require large amounts of unlabeled data and are expensive to train. In this work, we propose a task-specific structured pruning method CoFi (Coarse- and Fine-grained Pruning), which delivers highly parallelizable subnetworks and matches the distillation methods in both accuracy and latency, without resorting to any unlabeled data. Our key insight is to jointly prune coarse-grained (e.g., layers) and fine-grained (e.g., heads and hidden units) modules, which controls the pruning decision of each parameter with masks of different granularity. We also devise a layerwise distillation strategy to transfer knowledge from unpruned to pruned models during optimization. Our experiments on GLUE and SQuAD datasets show that CoFi yields models with over 10x speedups with a small accuracy drop, showing its effectiveness and efficiency compared to previous pruning and distillation approaches.
With the rapid development of deep learning, the increasing complexity and scale of parameters make training a new model increasingly resource-intensive. In this paper, we start from the classic convolutional neural network (CNN) and explore a paradigm that does not require training to obtain new models. Similar to the birth of CNN inspired by receptive fields in the biological visual system, we draw inspiration from the information subsystem pathways in the biological visual system and propose Model Disassembling and Assembling (MDA). During model disassembling, we introduce the concept of relative contribution and propose a component locating technique to extract task-aware components from trained CNN classifiers. For model assembling, we present the alignment padding strategy and parameter scaling strategy to construct a new model tailored for a specific task, utilizing the disassembled task-aware components. The entire process is akin to playing with LEGO bricks, enabling arbitrary assembly of new models, and providing a novel perspective for model creation and reuse. Extensive experiments showcase that task-aware components disassembled from CNN classifiers or new models assembled using these components closely match or even surpass the performance of the baseline, demonstrating its promising results for model reuse. Furthermore, MDA exhibits diverse potential applications, with comprehensive experiments exploring model decision route analysis, model compression, knowledge distillation, and more. The code is available at https://github.com/jiaconghu/Model-LEGO.