In recent years, deep learning has spread rapidly, and deeper, larger models have been proposed. However, the calculation cost becomes enormous as the size of the models becomes larger. Various techniques for compressing the size of the models have been proposed to improve performance while reducing computational costs. One of the methods to compress the size of the models is knowledge distillation (KD). Knowledge distillation is a technique for transferring knowledge of deep or ensemble models with many parameters (teacher model) to smaller shallow models (student model). Since the purpose of knowledge distillation is to increase the similarity between the teacher model and the student model, we propose to introduce the concept of metric learning into knowledge distillation to make the student model closer to the teacher model using pairs or triplets of the training samples. In metric learning, the researchers are developing the methods to build a model that can increase the similarity of outputs for similar samples. Metric learning aims at reducing the distance between similar and increasing the distance between dissimilar. The functionality of the metric learning to reduce the differences between similar outputs can be used for the knowledge distillation to reduce the differences between the outputs of the teacher model and the student model. Since the outputs of the teacher model for different objects are usually different, the student model needs to distinguish them. We think that metric learning can clarify the difference between the different outputs, and the performance of the student model could be improved. We have performed experiments to compare the proposed method with state-of-the-art knowledge distillation methods.
Institute of Computing Technology (ICT), Chinese Academy of Sciences (CAS)（中国科学院计算技术研究所）；University of Chinese Academy of Sciences（中国科学院大学）；Horizon Robotics（地平线机器人）；Peng Cheng Laboratory（鹏城实验室）
We study a series of recognition tasks in two realistic scenarios requiring the analysis of faces under strong occlusion. On the one hand, we aim to recognize facial expressions of people wearing Virtual Reality (VR) headsets. On the other hand, we aim to estimate the age and identify the gender of people wearing surgical masks. For all these tasks, the common ground is that half of the face is occluded. In this challenging setting, we show that convolutional neural networks (CNNs) trained on fully-visible faces exhibit very low performance levels. While fine-tuning the deep learning models on…
Novustech Services（诺沃斯特科技服务公司）；University of Bucharest（布加勒斯特大学）
In this paper, we study the task of facial expression recognition under strong occlusion. We are particularly interested in cases where 50% of the face is occluded, e.g. when the subject wears a Virtual Reality (VR) headset. While previous studies show that pre-training convolutional neural networks (CNNs) on fully-visible (non-occluded) faces improves the accuracy, we propose to employ knowledge distillation to achieve further improvements. First of all, we employ the classic teacher-student training strategy, in which the teacher is a CNN trained on fully-visible faces and the student is a C…
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…
Computer Vision Laboratory, University of Nottingham, UK（诺丁汉大学计算机视觉实验室）；Samsung AI Center, Cambridge, UK（三星人工智能中心剑桥）；Queen Mary University of London, London, UK（伦敦玛丽女王大学）
This paper addresses the problem of model compression via knowledge distillation. To this end, we propose a new knowledge distillation method based on transferring feature statistics, specifically the channel-wise mean and variance, from the teacher to the student. Our method goes beyond the standard way of enforcing the mean and variance of the student to be similar to those of the teacher through an $L_2$ loss, which we found it to be of limited effectiveness. Specifically, we propose a new loss based on adaptive instance normalization to effectively transfer the feature statistics. The main…
Huazhong University of Science and Technology（华中科技大学）
Most existing 3D object recognition algorithms focus on leveraging the strong discriminative power of deep learning models with softmax loss for the classification of 3D data, while learning discriminative features with deep metric learning for 3D object retrieval is more or less neglected. In the paper, we study variants of deep metric learning losses for 3D object retrieval, which did not receive enough attention from this area. First , two kinds of representative losses, triplet loss and center loss, are introduced which could learn more discriminative features than traditional classificati…
Knowledge distillation aims to transfer knowledge to the student model by utilizing the predictions/features of the teacher model, and feature-based distillation has recently shown its superiority over logit-based distillation. However, due to the cumbersome computation and storage of extra feature transformation, the training overhead of feature-based methods is much higher than that of logit-based distillation. In this work, we revisit the logit-based knowledge distillation, and observe that the existing logit-based distillation methods treat the prediction logits only in the instance level,…
Cooperative Medianet Innovation Center, Shanghai Jiao Tong University（上海交通大学协同媒体创新中心）
With the explosive growth of image databases, deep hashing, which learns compact binary descriptors for images, has become critical for fast image retrieval. Many existing deep hashing methods leverage quantization loss, defined as distance between the features before and after quantization, to reduce the error from binarizing features. While minimizing the quantization loss guarantees that quantization has minimal effect on retrieval accuracy, it unfortunately significantly reduces the expressiveness of features even before the quantization. In this paper, we show that the above definition of…
Enriching Knowledge Distillation with Intra-Class Contrastive Learning
Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China（新一代人工智能技术及其跨学科应用重点实验室（东南大学），教育部，中国）
针对知识蒸馏中教师软标签缺乏类内多样表示的问题，本文提出带 margin loss 的类内对比学习教师训练方法，并通过理论与实验验证其有效性。
University of Southern California（南加州大学）；Facebook AI
To reduce a model size but retain performance, we often rely on knowledge distillation (KD) which transfers knowledge from a large "teacher" model to a smaller "student" model. However, KD on multimodal datasets such as vision-language tasks is relatively unexplored, and digesting multimodal information is challenging since different modalities present different types of information. In this paper, we perform a large-scale empirical study to investigate the importance and effects of each modality in knowledge distillation. Furthermore, we introduce a multimodal knowledge distillation framework…
Amasya University（阿马西亚大学）；Bilkent University（比尔肯特大学）；National Magnetic Resonance Research Center, Bilkent University（比尔肯特大学国家磁共振研究中心）；Sabuncu Brain Research Center, Bilkent University（比尔肯特大学萨本朱脑研究中心）
Melanoma is a fatal skin cancer that is curable and has dramatically increasing survival rate when diagnosed at early stages. Learning-based methods hold significant promise for the detection of melanoma from dermoscopic images. However, since melanoma is a rare disease, existing databases of skin lesions predominantly contain highly imbalanced numbers of benign versus malignant samples. In turn, this imbalance introduces substantial bias in classification models due to the statistical dominance of the majority class. To address this issue, we introduce a deep clustering approach based on the …
Distance metric learning (DML) is to learn the embeddings where examples from the same class are closer than examples from different classes. It can be cast as an optimization problem with triplet constraints. Due to the vast number of triplet constraints, a sampling strategy is essential for DML. With the tremendous success of deep learning in classifications, it has been applied for DML. When learning embeddings with deep neural networks (DNNs), only a mini-batch of data is available at each iteration. The set of triplet constraints has to be sampled within the mini-batch. Since a mini-batch…
Convolutional neural networks (CNNs) have achieved a great success in face recognition, which unfortunately comes at the cost of massive computation and storage consumption. Many compact face recognition networks are thus proposed to resolve this problem. Triplet loss is effective to further improve the performance of those compact models. However, it normally employs a fixed margin to all the samples, which neglects the informative similarity structures between different identities. In this paper, we propose an enhanced version of triplet loss, named triplet distillation, which exploits the c…
Descriptor Distillation: a Teacher-Student-Regularized Framework for Learning Local Descriptors
State Key Laboratory of Multimodal Artificial Intelligence Systems, CASIA（中国科学院自动化研究所多模态人工智能系统国家重点实验室）；School of Artificial Intelligence, UCAS（中国科学院大学人工智能学院）
Triplet loss, one of the deep metric learning (DML) methods, is to learn the embeddings where examples from the same class are closer than examples from different classes. Motivated by DML, we propose an effective BP-Triplet Loss for unsupervised domain adaption (UDA) from the perspective of Bayesian learning and we name the model as BP-Triplet Net. In previous metric learning based methods for UDA, sample pairs across domains are treated equally, which is not appropriate due to the domain bias. In our work, considering the different importance of pair-wise samples for both feature learning an…
Beihang University（北京航空航天大学）；Research Institute of Beihang University in Shenzhen（北京航空航天大学深圳研究院）；Beijing Advanced Innovation Center for Big Data and Brain Computing（北京大数据与脑计算先进创新中心）
How to obtain the desirable representation of a 3D shape, which is discriminative across categories and polymerized within classes, is a significant challenge in 3D shape retrieval. Most existing 3D shape retrieval methods focus on capturing strong discriminative shape representation with softmax loss for the classification task, while the shape feature learning with metric loss is neglected for 3D shape retrieval. In this paper, we address this problem based on the intuition that the cosine distance of shape embeddings should be close enough within the same class and far away across categorie…
Seoul National University（首尔国立大学）；Hanyang University（汉阳大学）；NAVER Corp（NAVER公司）
An activation boundary for a neuron refers to a separating hyperplane that determines whether the neuron is activated or deactivated. It has been long considered in neural networks that the activations of neurons, rather than their exact output values, play the most important role in forming classification friendly partitions of the hidden feature space. However, as far as we know, this aspect of neural networks has not been considered in the literature of knowledge transfer. In this paper, we propose a knowledge transfer method via distillation of activation boundaries formed by hidden neuron…
Tsinghua Shenzhen International Graduate School（清华大学深圳国际研究生院）；International Digital Economy Academy (IDEA)（国际数字经济研究院（IDEA））
Knowledge Distillation (KD) has developed extensively and boosted various tasks. The classical KD method adds the KD loss to the original cross-entropy (CE) loss. We try to decompose the KD loss to explore its relation with the CE loss. Surprisingly, we find it can be regarded as a combination of the CE loss and an extra loss which has the identical form as the CE loss. However, we notice the extra loss forces the student's relative probability to learn the teacher's absolute probability. Moreover, the sum of the two probabilities is different, making it hard to optimize. To address this issue…
Yanshan University（燕山大学）；Hebei Normal University of Science and Technology（河北科技师范学院）
With the widespread use of touch-screen devices, it is more and more convenient for people to draw sketches on screen. This results in the demand for automatically understanding the sketches. Thus, the sketch recognition task becomes more significant than before. To accomplish this task, it is necessary to solve the critical issue of improving the distinction of the sketch features. To this end, we have made efforts in three aspects. First, a novel multi-scale residual block is designed. Compared with the conventional basic residual block, it can better perceive multi-scale information and red…
Zhejiang University（浙江大学）；Hangzhou City University（杭州城市大学）；Zhejiang Lab（之江实验室）
Most existing online knowledge distillation(OKD) techniques typically require sophisticated modules to produce diverse knowledge for improving students' generalization ability. In this paper, we strive to fully utilize multi-model settings instead of well-designed modules to achieve a distillation effect with excellent generalization performance. Generally, model generalization can be reflected in the flatness of the loss landscape. Since averaging parameters of multiple models can find flatter minima, we are inspired to extend the process to the sampled convex combinations of multi-student mo…
From Knowledge Distillation to Self-Knowledge Distillation: A Unified Approach with Normalized Loss and Customized Soft Labels
Tsinghua Shenzhen International Graduate School（清华大学深圳国际研究生院）；International Digital Economy Academy (IDEA)（国际数字经济研究院）；Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）
Despite the empirical success of knowledge distillation, current state-of-the-art methods are computationally expensive to train, which makes them difficult to adopt in practice. To address this problem, we introduce two distinct complementary losses inspired by a cheap entropy-like estimator. These losses aim to maximise the correlation and mutual information between the student and teacher representations. Our method incurs significantly less training overheads than other approaches and achieves competitive performance to the state-of-the-art on the knowledge distillation and cross-model tra…
Indian Institute of Technology Mandi（印度理工学院曼迪分校）；Indian Institute of Technology Indore（印度理工学院印多尔分校）
Designing an end-to-end deep learning network to match the biometric features with limited training samples is an extremely challenging task. To address this problem, we propose a new way to design an end-to-end deep CNN framework i.e., PVSNet that works in two major steps: first, an encoder-decoder network is used to learn generative domain-specific features followed by a Siamese network in which convolutional layers are pre-trained in an unsupervised fashion as an autoencoder. The proposed model is trained via triplet loss function that is adjusted for learning feature embeddings in a way th…
Towards Lossless Dataset Distillation via Difficulty-Aligned Trajectory Matching
National University of Singapore（新加坡国立大学）；Massachusetts Institute of Technology（麻省理工学院）；Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）；Xidian University（西安电子科技大学）
Hashing method maps similar data to binary hashcodes with smaller hamming distance, which has received a broad attention due to its low storage cost and fast retrieval speed. With the rapid development of deep learning, deep hashing methods have achieved promising results in efficient information retrieval. Most of the existing deep hashing methods adopt pairwise or triplet losses to deal with similarities underlying the data, but the training is difficult and less efficient because $O(n^2)$ data pairs and $O(n^3)$ triplets are involved. To address these issues, we propose a novel deep hashing…
Knowledge Distillation (KD) is a strategy for the definition of a set of transferability gangways to improve the efficiency of Convolutional Neural Networks. Feature-based Knowledge Distillation is a subfield of KD that relies on intermediate network representations, either unaltered or depth-reduced via maximum activation maps, as the source knowledge. In this paper, we propose and analyse the use of a 2D frequency transform of the activation maps before transferring them. We pose that\textemdash by using global image cues rather than pixel estimates, this strategy enhances knowledge transfer…
Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）；University of Chinese Academy of Sciences（中国科学院大学）；Nanjing University of Science and Technology（南京理工大学）
Deep models trained on long-tailed datasets exhibit unsatisfactory performance on tail classes. Existing methods usually modify the classification loss to increase the learning focus on tail classes, which unexpectedly sacrifice the performance on head classes. In fact, this scheme leads to a contradiction between the two goals of long-tailed learning, i.e., learning generalizable representations and facilitating learning for tail classes. In this work, we explore knowledge distillation in long-tailed scenarios and propose a novel distillation framework, named Balanced Knowledge Distillation (…
In this work, we address the issues of missing modalities that have arisen from the Visual Question Answer-Difference prediction task and find a novel method to solve the task at hand. We address the missing modality-the ground truth answers-that are not present at test time and use a privileged knowledge distillation scheme to deal with the issue of the missing modality. In order to efficiently do so, we first introduce a model, the "Big" Teacher, that takes the image/question/answer triplet as its input and outperforms the baseline, then use a combination of models to distill knowledge to a …
University of Science and Technology of China（中国科学技术大学）；Sun Yat-sen University（中山大学）
Knowledge distillation is to transfer the knowledge from the data learned by the teacher network to the student network, so that the student has the advantage of less parameters and less calculations, and the accuracy is close to the teacher. In this paper, we propose a new distillation method, which contains two transfer distillation strategies and a loss decay strategy. The first transfer strategy is based on channel-wise attention, called Channel Distillation (CD). CD transfers the channel information from the teacher to the student. The second is Guided Knowledge Distillation (GKD). Unlike…
We propose a simple modification from a fixed margin triplet loss to an adaptive margin triplet loss. While the original triplet loss is used widely in classification problems such as face recognition, face re-identification and fine-grained similarity, our proposed loss is well suited for rating datasets in which the ratings are continuous values. In contrast to original triplet loss where we have to sample data carefully, in out method, we can generate triplets using the whole dataset, and the optimization can still converge without frequently running into a model collapsing issue. The adapt…
