Semantic-aware knowledge distillation with parameter-free feature uniformization

Knowledge Distillation (KD) aims to distill the dark knowledge of a high-powered teacher network into a student network, which can improve the capacity of student network and has been successfully applied to semantic segmentation. However, the standard knowledge distillation approaches merely represent the supervisory signal of teacher network as the dark knowledge, while ignoring the impact of network architecture during distillation. In this paper, we found that the student network with a more similar architecture against the teacher network obtains more performance gain from distillation. Therefore, a more generalized paradigm for knowledge distillation is to distill both the soft-label and the structure of the teacher network. We propose a novel Structural Distillation (SD) method which introduces the structural similarity constraints into vanilla knowledge distillation. We leverage Neural Architecture Search technique to search optimal student structure for semantic segmentation from a well-designed search space, which mimics the given teacher both in terms of soft-label and network structure. Experiment results make clear that our proposed method outperforms both the NAS with conventional Knowledge Distillation and human-designed methods, and achieves sota performance on the Cityscapes dataset under various platform-aware latency constraints. Furthermore, the best architecture discovered on Cityscapes also transfers well to the PASCAL VOC2012 dataset.

This paper presents a Cosine Similarity-Based Knowledge Distillation (CSKD) for robust, lightweight object detectors. Knowledge Distillation (KD) has been effective in enhancing the performance of compact models in image classification by leveraging deep CNN models. However, the complex and multifaceted nature of object detection, characterized by its modular design and multitasking requirements, poses significant challenges for traditional KD techniques. These challenges are further compounded by the conventional reliance on the Mean Squared Error (MSE) loss function and the limited application of enhanced feature representations to the training phase. Addressing these limitations, the proposed CSKD method combines cosine similarity guidance with MSE loss to facilitate a more effective knowledge transfer from the teacher model to the student model. This is achieved by distilling both intermediate features and prediction outputs, aided by an assistant prediction branch designed to learn directly from the teacher’s predictions. This dual-faceted distillation strategy enables the student model to better mimic the teacher model’s behavior, leading to improved performance. The proposed method demonstrates versatility and robustness across various object detector architectures without the need for additional feature enhancement layers during training. Notably, employing ResNet-50 as the teacher model and ResNet-18 as the student model, we achieve new benchmarks in KD for object detection across several architectures, including Faster-RCNN, RetinaNet, FCOS, and GFL, with respective mAP scores of 36.6, 35.2, 35.9, and 38.9. These results highlights the effectiveness of CSKD in advancing the state-of-the-art in KD for object detection, offering a compelling solution to the challenges previously faced by traditional KD methods in this domain. The code of the proposed CSKD is available at https://github.com/swkdn16/CSKD.

Knowledge distillation is a widely used neural network model compression technique. In general, the knowledge distillation transfer the knowledge from a large pre-trained teacher network with superior performance to a small student network enables the student network to achieve better performance. This paper proposes a triplet knowledge distillation framework (abbreviated as TKD), which introduces a smaller assistant network into the knowledge distillation structure. The performance of the assistant network is lower than that of the student network. During the training of the TKD, by minimizing the Mean Squared Error(MSE) loss function, the output of the student network will closer to the output of the teacher network and further from that of the assistant network. Therefore, the student network can learn more expressive knowledge from the teacher network while throwing away mistaken knowledge in the assistant network. Finally, the student network achieves a surprising performance even superior to the teacher network. We have demonstrated the effectiveness of TKD by extensive experiments on benchmark datasets(CIFAR-10, CIFAR-100, SVHN, STL-10). When using VGGNet as an experimental model, the student network VGGNet13 achieving 94.29%, 75.30%, 95.53%, and 87.61% accuracy on the CIFAR-10, CIFAR-100, SVHN, and STL-10 datasets, improved by 1.24%, 2.81%, 0.40%, and 2.32%, respectively.

Knowledge distillation is a deep learning method that mimics the way that humans teach, i.e., a teacher network is used to guide the training of a student one. Knowledge distillation can generate an efficient student network to facilitate deployment in resource-constrained edge computing devices. Existing studies have typically mined knowledge from a teacher network and transferred it to a student one. The latter can only passively receive knowledge but cannot understand how the former acquires the knowledge, thus limiting the latter’s performance improvement. Inspired by the old Chinese saying “Give a man a fish and you feed him for a day; teach a man how to fish and you feed him for a lifetime,” this work proposes a Skill-transferring Knowledge Distillation (SKD) method to boost a student network’s ability to create new valuable knowledge. SKD consists of two main meta-learning networks: Teacher Behavior Teaching and Teacher Experience Teaching. The former captures the process of a teacher network’s learning behavior in the hidden layers and can predict the teacher network’s subsequent behavior based on previous ones. The latter models the optimal empirical knowledge of a teacher network’s output layer at each learning stage. With their help, a teacher network can provide its actions to a student one in the subsequent behavior and its optimal empirical knowledge in the current stage. The overall performance of SKD is verified through its application to multiple object recognition tasks and comparison with the state of the art.

Knowledge distillation is an effective approach to transferring knowledge from the large teacher network to its small proxy student one, thereby letting the proxy student work on those resource-limited mobile devices. Most previous arts manually select the paired intermediate layers of teacher and student networks to align their pertinent features by dimension reduction. This sort of approach may confront information loss and insufficient layer-wise alignment that limit knowledge transferability. In this paper, we propose a simple and effective knowledge distillation method named attentive similarity matching (ASM). ASM at first concatenates the teacher’s intermediate features and the student’s ones together to enhance similarity representation of all the student’s layers, without involving dimension reduction, then align all cross-layer advanced similarities in an attentively weighted manner for semantic calibration. Experiments of image classification on three popular datasets show the effectiveness of the proposed method as compared to its previous cousins.

Knowledge distillation (KD) is an effective model compression technique where a compact student network is taught to mimic the behavior of a complex and highly trained teacher network. In contrast, Mutual Learning (ML) provides an alternative strategy where multiple simple student networks benefit from sharing knowledge, even in the absence of a powerful but static teacher network. Motivated by these findings, we propose a single-teacher, multi-student framework that leverages both KD and ML to achieve better performance. Furthermore, an online distillation strategy is utilized to train the teacher and students simultaneously. To evaluate the performance of the proposed approach, extensive experiments were conducted using three different versions of teacher-student networks on benchmark biomedical classification (MSI vs. MSS) and object detection (Polyp Detection) tasks. Ensemble of student networks trained in the proposed manner achieved better results than the ensemble of students trained using KD or ML individually, establishing the benefit of augmenting knowledge transfer from teacher to students with peer-to-peer learning between students.

A tucker decomposition based knowledge distillation for intelligent edge applications

Deep learning techniques have recently demonstrated remarkable success in numerous domains. Typically, the success of these deep learning models is measured in terms of performance metrics such as accuracy and mean average precision (mAP). Generally, a model's high performance is highly valued, but it frequently comes at the expense of substantial energy costs and carbon footprint emissions during the model building step. Massive emission of CO2 has a deleterious impact on life on earth in general and is a serious ethical concern that is largely ignored in deep learning research. In this article, we mainly focus on environmental costs and the means of mitigating carbon footprints in deep learning models, with a particular focus on models created using knowledge distillation (KD). Deep learning models typically contain a large number of parameters, resulting in a 'heavy' model. A heavy model scores high on performance metrics but is incompatible with mobile and edge computing devices. Model compression techniques such as knowledge distillation enable the creation of lightweight, deployable models for these low-resource devices. KD generates lighter models and typically performs with slightly less accuracy than the heavier teacher model (model accuracy by the teacher model on CIFAR 10, CIFAR 100, and TinyImageNet is 95.04%, 76.03%, and 63.39%; model accuracy by KD is 91.78%, 69.7%, and 60.49%). Although the distillation process makes models deployable on low-resource devices, they were found to consume an exorbitant amount of energy and have a substantial carbon footprint (15.8, 17.9, and 13.5 times more carbon compared to the corresponding teacher model). The enormous environmental cost is primarily attributable to the tuning of the hyperparameter, Temperature (τ). In this article, we propose measuring the environmental costs of deep learning work (in terms of GFLOPS in millions, energy consumption in kWh, and CO2 equivalent in grams). In order to create lightweight models with low environmental costs, we propose a straightforward yet effective method for selecting a hyperparameter (τ) using a stochastic approach for each training batch fed into the models. We applied knowledge distillation (including its data-free variant) to problems involving image classification and object detection. To evaluate the robustness of our method, we ran experiments on various datasets (CIFAR 10, CIFAR 100, Tiny ImageNet, and PASCAL VOC) and models (ResNet18, MobileNetV2, Wrn-40-2). Our novel approach reduces the environmental costs by a large margin by eliminating the requirement of expensive hyperparameter tuning without sacrificing performance. Empirical results on the CIFAR 10 dataset show that the stochastic technique achieves an accuracy of 91.67%, whereas tuning achieves an accuracy of 91.78%-however, the stochastic approach reduces the energy consumption and CO2 equivalent each by a factor of 19. Similar results have been obtained with CIFAR 100 and TinyImageNet dataset. This pattern is also observed in object detection classification on the PASCAL VOC dataset, where the tuning technique performs similarly to the stochastic technique, with a difference of 0.03% mAP favoring the stochastic technique while reducing the energy consumptions and CO2 emission each by a factor of 18.5.

In this work, we consider transferring the structure information from large networks to compact ones for dense prediction tasks in computer vision. Previous knowledge distillation strategies used for dense prediction tasks often directly borrow the distillation scheme for image classification and perform knowledge distillation for each pixel separately, leading to sub-optimal performance. Here we propose to distill structured knowledge from large networks to compact networks, taking into account the fact that dense prediction is a structured prediction problem. Specifically, we study two structured distillation schemes: i) pair-wise distillation that distills the pair-wise similarities by building a static graph; and ii) holistic distillation that uses adversarial training to distill holistic knowledge. The effectiveness of our knowledge distillation approaches is demonstrated by experiments on three dense prediction tasks: semantic segmentation, depth estimation and object detection. Code is available at https://git.io/StructKD.

While pursuing high performance, most currently proposed object detection algorithms are composed of huge and complex neural network and large amount of parameters. As a results, many of them can not be implemented on edge devices due to the limitation of computation resources. To address this, in this paper, we designed a model compression scheme for object detection algorithm based on knowledge distillation. Firstly, a novel loss function based on mutual information have been proposed for similarity measurement between the teacher network and the student network. Then, a new knowledge distillation scheme is proposed for object detection network. The proposed algorithm has been verified on PASACL VOC dataset. The experimental results have demonstrated the effectiveness of our method.

A single model usually cannot learn all the appropriate features with limited data, thus leading to poor performance when test data are used. To improve model performance, we propose a teacher-student collaborative knowledge distillation (TSKD) method based on knowledge distillation and self-distillation. The method consists of two parts: learning in the teacher network and self-teaching in the student network. Learning in the teacher network allows the student network to use knowledge from the teacher network. Self-teaching in the student network is to build a multi-exit network based on self-distillation and provide deep features as supervised information for training. In the inference stage, we use ensembles to vote on the classification results of multiple sub-models in the student network. The experimental results demonstrate the superior performance of our method compared with a traditional knowledge distillation method and a self-distillation-based multi-exit network.

We present Self-Guidance, a simple way to train deep neural networks via knowledge distillation. The basic idea is to train sub-network to match the prediction of the full network, so-called "Self-Guidance". Under the "teacher-student" framework, we construct both teacher and student within the same target network. Student network is the sub-networks that randomly skip some portions of the full network. The teacher network is the full network, can be considered as the ensemble of all possible student networks. The training process is performed in a closed-loop: (1) Forward prediction contains two passes that generate student and teacher predictions. (2) Backward distillation allows knowledge transfer from the teacher back to students. Comprehensive evaluations show that our approach improves the generalization ability of deep neural networks to a significant margin. The results prove our superior performance in both image classification on CIFAR10, CIFAR100, and facial expression recognition on FER-2013 and RAF.

Automated multi-label chest X-rays (CXR) image classification has achieved substantial progress in clinical diagnosis via utilizing sophisticated deep learning approaches. However, most deep models have high computational demands, which makes them less feasible for compact devices with low computational requirements. To overcome this problem, we propose a knowledge distillation (KD) strategy to create the compact deep learning model for the real-time multi-label CXR image classification. We study different alternatives of CNNs and Transforms as the teacher to distill the knowledge to a smaller student. Then, we employed explainable artificial intelligence (XAI) to provide the visual explanation for the model decision improved by the KD. Our results on three benchmark CXR datasets show that our KD strategy provides the improved performance on the compact student model, thus being the feasible choice for many limited hardware platforms. For instance, when using DenseNet161 as the teacher network, EEEA-Net-C2 achieved an AUC of 83.7%, 87.1%, and 88.7% on the ChestX-ray14, CheXpert, and PadChest datasets, respectively, with fewer parameters of 4.7 million and computational cost of 0.3 billion FLOPS.

Numerous transformer-based medical image segmentation methods have been proposed and achieved good segmentation results. However, it is still a challenge to train and deploy transformer networks to mobile medical devices due to a large number of model parameters. To resolve the training and model parameter problems, in this paper, we propose a Transformer-based network for Medical Image Segmentation using Knowledge Distillation named MISTKD. The MISTKD consists of a teacher network and a student network. It achieves comparable performance to state-of-the-art transformer works using fewer parameters by employing the teacher network to train the student network. The training can be implemented by extracting the sequence in the teacher and student encoder networks during the training procedure. The losses between sequences are further calculated, thus the student network can learn from the teacher network. The experimental results on Synapse show that the proposed work achieves competitive performance using only one-eighth parameters.

Knowledge distillation, in which the parameter values learned in a large teacher network are transferred to a smaller student network, is a popular and effective network compression method. Recently, researchers have proposed methods to improve the performance of a student network by using a Generative Adverserial Network (GAN). However, because a GAN is an architecture that is ideally used to create realistic synthetic images, a pure GAN architecture may not be ideally suited for knowledge distillation. In knowledge distillation for image signal processing, synthetic images do not need to be realistic, but instead should include features that help the training of the student network. In the proposed Generative Image Processing (GIP) method, this is accomplished by using only the generator portion of a GAN and utilizing special techniques to capture the distinguishing feature capability of the teacher network. Experimental results show that the GIP method outperforms knowledge distillation using GANs as well as training using only knowledge distillation.