Compared to single-modal knowledge distillation, cross-modal knowledge distillation faces more severe challenges due to domain gaps between modalities. Although various methods have proposed various solutions to overcome these challenges, there is still limited research on how domain gaps affect cross-modal knowledge distillation. This paper provides an in-depth analysis and evaluation of this issue. We first introduce the Non-Target Divergence Hypothesis (NTDH) to reveal the impact of domain gaps on cross-modal knowledge distillation. Our key finding is that domain gaps between modalities lead to distribution differences in non-target classes, and the smaller these differences, the better the performance of cross-modal knowledge distillation. Subsequently, based on Vapnik-Chervonenkis (VC) theory, we derive the upper and lower bounds of the approximation error for cross-modal knowledge distillation, thereby theoretically validating the NTDH. Finally, experiments on five cross-modal datasets further confirm the validity, generalisability, and applicability of the NTDH.
School of Computer Science and Engineering, Central South University（中南大学计算机科学与工程学院）；Institute of Artificial Intelligence, Shaoxing University（绍兴市文理学院人工智能学院）；National University of Defense Technology（国防科技大学）；Cyberspace Institute of Advanced Technology, Guangzhou University（广州大学网络空间先进技术研究院）
Knowledge distillation (KD), transferring knowledge from a cumbersome teacher model to a lightweight student model, has been investigated to design efficient neural architectures. Generally, the objective function of KD is the Kullback-Leibler (KL) divergence loss between the softened probability distributions of the teacher model and the student model with the temperature scaling hyperparameter tau. Despite its widespread use, few studies have discussed the influence of such softening on generalization. Here, we theoretically show that the KL divergence loss focuses on the logit matching when…
Shanghai Em-Data Technology Co.（上海翼眸科技有限公司）；The University of Adelaide（阿德莱德大学）；Monash University（莫纳什大学）；Baidu Inc.（百度公司）
Knowledge distillation (KD) has been proven to be a simple and effective tool for training compact models. Almost all KD variants for dense prediction tasks align the student and teacher networks' feature maps in the spatial domain, typically by minimizing point-wise and/or pair-wise discrepancy. Observing that in semantic segmentation, some layers' feature activations of each channel tend to encode saliency of scene categories (analogue to class activation mapping), we propose to align features channel-wise between the student and teacher networks. To this end, we first transform the feature …
One-for-All: Bridge the Gap Between Heterogeneous Architectures in Knowledge Distillation
School of information and Electronics, Beijing Institute of Technology（北京理工大学信息与电子学院）；Huawei Noah’s Ark Lab（华为诺亚方舟实验室）；School of Computer Science, Faculty of Engineering, The University of Sydney（悉尼大学工程学院计算机科学学院）
Adaptive Dual-Teacher Distillation with Subnetwork Rectification for Bridging Semantic Gaps in Black-Box Domain Adaptation
School of Computer Science and Engineering, Tianjin University of Technology（天津理工大学计算机科学与工程学院）；School of Artificial Intelligence, Tianjin University（天津大学人工智能学院）
Clemson University（克莱姆森大学）；University at Buffalo（布法罗大学）
Knowledge distillation (KD) has become an important technique for model compression and knowledge transfer. In this work, we first perform a comprehensive analysis of the knowledge transferred by different KD methods. We demonstrate that traditional KD methods, which minimize the KL divergence of softmax outputs between networks, are related to the knowledge alignment of an individual sample only. Meanwhile, recent contrastive learning-based KD methods mainly transfer relational knowledge between different samples, namely, knowledge correlation. While it is important to transfer the full knowl…
Hangzhou Normal University（杭州师范大学）；Zhejiang University（浙江大学）
Existing state-of-the-art human pose estimation methods require heavy computational resources for accurate predictions. One promising technique to obtain an accurate yet lightweight pose estimator is knowledge distillation, which distills the pose knowledge from a powerful teacher model to a less-parameterized student model. However, existing pose distillation works rely on a heavy pre-trained estimator to perform knowledge transfer and require a complex two-stage learning procedure. In this work, we investigate a novel Online Knowledge Distillation framework by distilling Human Pose structure…
ShanghaiTech University（上海科技大学）；JD Explore Academy（京东探索研究院）；Meituan Inc.（美团公司）；The University of Sydney（悉尼大学）；Shanghai Engineering Research Center of Intelligent Vision and Imaging（上海智能视觉与成像工程技术研究中心）；Shanghai Engineering Research Center of Energy Efficient and Custom AI IC（上海能效与定制人工智能集成电路工程技术研究中心）
Transformers are successfully applied to computer vision due to their powerful modeling capacity with self-attention. However, the excellent performance of transformers heavily depends on enormous training images. Thus, a data-efficient transformer solution is urgently needed. In this work, we propose an early knowledge distillation framework, which is termed as DearKD, to improve the data efficiency required by transformers. Our DearKD is a two-stage framework that first distills the inductive biases from the early intermediate layers of a CNN and then gives the transformer full play by train…
State Key Laboratory of Blockchain and Data Security, Zhejiang University（区块链与数据安全国家重点实验室，浙江大学）；School of Software Technology, Zhejiang University（浙江大学软件技术学院）；Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security（杭州高新技术区（滨江区）区块链与数据安全研究院）；University at Buffalo, State University of New York（纽约州立大学布法罗分校）；NingboTech University（宁波科技学院）
Hefei University of Technology（合肥工业大学）；University of Science and Technology of China（中国科学技术大学）；Nanyang Technological University（南洋理工大学）；The Chinese University of Hong Kong（香港中文大学）；The University of Hong Kong（香港大学）；Harbin Institute of Technology, Shenzhen（哈尔滨工业大学（深圳））
Deep Learning has a hierarchical network architecture to represent the complicated feature of input patterns. We have developed the adaptive structure learning method of Deep Belief Network (DBN) that can discover an optimal number of hidden neurons for given input data in a Restricted Boltzmann Machine (RBM) by neuron generation-annihilation algorithm, and can obtain the appropriate number of hidden layers in DBN. In this paper, our model is applied to a facial expression image data set, AffectNet. The system has higher classification capability than the traditional CNN. However, our model wa…
The primary goal of knowledge distillation (KD) is to encapsulate the information of a model learned from a teacher network into a student network, with the latter being more compact than the former. Existing work, e.g., using Kullback-Leibler divergence for distillation, may fail to capture important structural knowledge in the teacher network and often lacks the ability for feature generalization, particularly in situations when teacher and student are built to address different classification tasks. We propose Wasserstein Contrastive Representation Distillation (WCoRD), which leverages both…
Samsung Research America（三星美国研究院）；Georgia Institute of Technology（佐治亚理工学院）
Knowledge distillation (KD) is a substantial strategy for transferring learned knowledge from one neural network model to another. A vast number of methods have been developed for this strategy. While most method designs a more efficient way to facilitate knowledge transfer, less attention has been put on comparing the effect of knowledge sources such as features, logits, and gradients. This work provides a new perspective to motivate a set of knowledge distillation strategies by approximating the classical KL-divergence criteria with different knowledge sources, making a systematic comparison…
The outpouring of various pre-trained models empowers knowledge distillation by providing abundant teacher resources, but there lacks a developed mechanism to utilize these teachers adequately. With a massive model repository composed of teachers pre-trained on diverse tasks, we must surmount two obstacles when using knowledge distillation to learn a new task. First, given a fixed computing budget, it is not affordable to try each teacher and train the student repeatedly, making it necessary to seek out the most contributive teacher precisely and efficiently. Second, semantic gaps exist betwee…
MIT CSAIL（麻省理工学院计算机科学与人工智能实验室）；Google Research（谷歌研究院）
Often we wish to transfer representational knowledge from one neural network to another. Examples include distilling a large network into a smaller one, transferring knowledge from one sensory modality to a second, or ensembling a collection of models into a single estimator. Knowledge distillation, the standard approach to these problems, minimizes the KL divergence between the probabilistic outputs of a teacher and student network. We demonstrate that this objective ignores important structural knowledge of the teacher network. This motivates an alternative objective by which we train a stud…
The performance of federated learning in neural networks is generally influenced by the heterogeneity of the data distribution. For a well-performing global model, taking a weighted average of the local models, as done by most existing federated learning algorithms, may not guarantee consistency with local models in the space of neural network maps. In this paper, we propose a novel framework of federated learning equipped with the process of decentralized knowledge distillation (FedDKD) (i.e., without data on the server). The FedDKD introduces a module of decentralized knowledge distillation …
Imperial College London（伦敦帝国理工学院）；The Chinese University of Hong Kong（香港中文大学）
Multi-modal learning is typically performed with network architectures containing modality-specific layers and shared layers, utilizing co-registered images of different modalities. We propose a novel learning scheme for unpaired cross-modality image segmentation, with a highly compact architecture achieving superior segmentation accuracy. In our method, we heavily reuse network parameters, by sharing all convolutional kernels across CT and MRI, and only employ modality-specific internal normalization layers which compute respective statistics. To effectively train such a highly compact model,…
In real applications, different computation-resource devices need different-depth networks (e.g., ResNet-18/34/50) with high-accuracy. Usually, existing methods either design multiple networks and train them independently, or construct depth-level/width-level dynamic neural networks which is hard to prove the accuracy of each sub-net. In this article, we propose an elegant Depth-Level Dynamic Neural Network (DDNN) integrated different-depth sub-nets of similar architectures. To improve the generalization of sub-nets, we design the Embedded-Knowledge-Distillation (EKD) training mechanism for th…
Facebook（脸书）；The University of Texas at Austin（德克萨斯大学奥斯汀分校）
Weight-sharing neural architecture search (NAS) is an effective technique for automating efficient neural architecture design. Weight-sharing NAS builds a supernet that assembles all the architectures as its sub-networks and jointly trains the supernet with the sub-networks. The success of weight-sharing NAS heavily relies on distilling the knowledge of the supernet to the sub-networks. However, we find that the widely used distillation divergence, i.e., KL divergence, may lead to student sub-networks that over-estimate or under-estimate the uncertainty of the teacher supernet, leading to infe…
HAD: Hierarchical Asymmetric Distillation to Bridge Spatio-Temporal Gaps in Event-Based Object Tracking
Sanya Science and Education Innovation Park, Wuhan University of Technology（武汉理工大学三亚科学教育创新园）；Hubei Key Laboratory of Transportation Internet of Things, School of Computer Science and Artificial Intelligence, Wuhan University of Technology（湖北省交通运输物联网重点实验室，计算机科学与人工智能学院，武汉理工大学）；State Key Laboratory for Multimedia Information Processing, Peking University（多媒体信息处理国家重点实验室，北京大学）
Self-supervised Normality Learning and Divergence Vector-guided Model Merging for Zero-shot Congenital Heart Disease Detection in Fetal Ultrasound Videos
University of Oxford（牛津大学）；University of Technology Nuremberg（纽伦堡工业大学）
Several methods of knowledge distillation have been developed for neural network compression. While they all use the KL divergence loss to align the soft outputs of the student model more closely with that of the teacher, the various methods differ in how the intermediate features of the student are encouraged to match those of the teacher. In this paper, we propose a simple-to-implement method using auxiliary classifiers at intermediate layers for matching features, which we refer to as multi-head knowledge distillation (MHKD). We add loss terms for training the student that measure the dissi…
Distribution Matching Distillation without Fake Score Network
Department of Computer Science and Engineering, Seoul National University（首尔国立大学计算机科学与工程系）；Department of Computer Science and Engineering, Sogang University（成均馆大学计算机科学与工程系）
To boost the performance, deep neural networks require deeper or wider network structures that involve massive computational and memory costs. To alleviate this issue, the self-knowledge distillation method regularizes the model by distilling the internal knowledge of the model itself. Conventional self-knowledge distillation methods require additional trainable parameters or are dependent on the data. In this paper, we propose a simple and effective self-knowledge distillation method using a dropout (SD-Dropout). SD-Dropout distills the posterior distributions of multiple models through a dro…
Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）；Beijing University of Posts and Telecommunications（北京邮电大学）；Shandong University of Science and Technology（山东科技大学）
Efficient medical image segmentation aims to provide accurate pixel-wise predictions for medical images with a lightweight implementation framework. However, lightweight frameworks generally fail to achieve superior performance and suffer from poor generalizable ability on cross-domain tasks. In this paper, we explore the generalizable knowledge distillation for the efficient segmentation of cross-domain medical images. Considering the domain gaps between different medical datasets, we propose the Model-Specific Alignment Networks (MSAN) to obtain the domain-invariant representations. Meanwhil…
Among the neural network compression techniques, knowledge distillation is an effective one which forces a simpler student network to mimic the output of a larger teacher network. However, most of such model distillation methods focus on the image-level classification task. Directly adapting these methods to the task of semantic segmentation only brings marginal improvements. In this paper, we propose a simple, yet effective knowledge representation referred to as pixel-wise feature similarities (PFS) to tackle the challenging distillation problem of semantic segmentation. The developed PFS en…
