With the swift advancement of deep learning, state-of-the-art algorithms have been utilized in various social situations. Nonetheless, some algorithms have been discovered to exhibit biases and provide unequal results. The current debiasing methods face challenges such as poor utilization of data or intricate training requirements. In this work, we found that the backdoor attack can construct an artificial bias similar to the model bias derived in standard training. Considering the strong adjustability of backdoor triggers, we are motivated to mitigate the model bias by carefully designing reverse artificial bias created from backdoor attack. Based on this, we propose a backdoor debiasing framework based on knowledge distillation, which effectively reduces the model bias from original data and minimizes security risks from the backdoor attack. The proposed solution is validated on both image and structured datasets, showing promising results. This work advances the understanding of backdoor attacks and highlights its potential for beneficial applications. The code for the study can be found at \url{https://anonymous.4open.science/r/DwB-BC07/}.
In this paper, we propose a language-universal adapter learning framework based on a pre-trained model for end-to-end multilingual automatic speech recognition (ASR). For acoustic modeling, the wav2vec 2.0 pre-trained model is fine-tuned by inserting language-specific and language-universal adapters. An online knowledge distillation is then used to enable the language-universal adapters to learn both language-specific and universal features. The linguistic information confusion is also reduced by leveraging language identifiers (LIDs). With LIDs we perform a position-wise modification on the multi-head attention outputs. In the inference procedure, the language-specific adapters are removed while the language-universal adapters are kept activated. The proposed method improves the recognition accuracy and addresses the linear increase of the number of adapters' parameters with the number of languages in common multilingual ASR systems. Experiments on the BABEL dataset confirm the effectiveness of the proposed framework. Compared to the conventional multilingual model, a 3.3% absolute error rate reduction is achieved. The code is available at: https://github.com/shen9712/UniversalAdapterLearning.
In this paper, we tackle the new task of video-based Activated Muscle Group Estimation (AMGE) aiming at identifying active muscle regions during physical activity in the wild. To this intent, we provide the MuscleMap dataset featuring >15K video clips with 135 different activities and 20 labeled muscle groups. This dataset opens the vistas to multiple video-based applications in sports and rehabilitation medicine under flexible environment constraints. The proposed MuscleMap dataset is constructed with YouTube videos, specifically targeting High-Intensity Interval Training (HIIT) physical exercise in the wild. To make the AMGE model applicable in real-life situations, it is crucial to ensure that the model can generalize well to numerous types of physical activities not present during training and involving new combinations of activated muscles. To achieve this, our benchmark also covers an evaluation setting where the model is exposed to activity types excluded from the training set. Our experiments reveal that the generalizability of existing architectures adapted for the AMGE task remains a challenge. Therefore, we also propose a new approach, TransM3E, which employs a multi-modality feature fusion mechanism between both the video transformer model and the skeleton-based graph convolution model with novel cross-modal knowledge distillation executed on multi-classification tokens. The proposed method surpasses all popular video classification models when dealing with both, previously seen and new types of physical activities. The database and code can be found at https://github.com/KPeng9510/MuscleMap.
Recent recommender systems have shown remarkable performance by using an ensemble of heterogeneous models. However, it is exceedingly costly because it requires resources and inference latency proportional to the number of models, which remains the bottleneck for production. Our work aims to transfer the ensemble knowledge of heterogeneous teachers to a lightweight student model using knowledge distillation (KD), to reduce the huge inference costs while retaining high accuracy. Through an empirical study, we find that the efficacy of distillation severely drops when transferring knowledge from heterogeneous teachers. Nevertheless, we show that an important signal to ease the difficulty can be obtained from the teacher's training trajectory. This paper proposes a new KD framework, named HetComp, that guides the student model by transferring easy-to-hard sequences of knowledge generated from the teachers' trajectories. To provide guidance according to the student's learning state, HetComp uses dynamic knowledge construction to provide progressively difficult ranking knowledge and adaptive knowledge transfer to gradually transfer finer-grained ranking information. Our comprehensive experiments show that HetComp significantly improves the distillation quality and the generalization of the student model.
Graph, such as citation networks, social networks, and transportation networks, are prevalent in the real world. Graph Neural Networks (GNNs) have gained widespread attention for their robust expressiveness and exceptional performance in various graph applications. However, the efficacy of GNNs is heavily reliant on sufficient data labels and complex network models, with the former obtaining hardly and the latter computing costly. To address the labeled data scarcity and high complexity of GNNs, Knowledge Distillation (KD) has been introduced to enhance existing GNNs. This technique involves transferring the soft-label supervision of the large teacher model to the small student model while maintaining prediction performance. This survey offers a comprehensive overview of Graph-based Knowledge Distillation methods, systematically categorizing and summarizing them while discussing their limitations and future directions. This paper first introduces the background of graph and KD. It then provides a comprehensive summary of three types of Graph-based Knowledge Distillation methods, namely Graph-based Knowledge Distillation for deep neural networks (DKD), Graph-based Knowledge Distillation for GNNs (GKD), and Self-Knowledge Distillation based Graph-based Knowledge Distillation (SKD). Each type is further divided into knowledge distillation methods based on the output layer, middle layer, and constructed graph. Subsequently, various algorithms' ideas are analyzed and compared, concluding with the advantages and disadvantages of each algorithm supported by experimental results. In addition, the applications of graph-based knowledge distillation in CV, NLP, RS, and other fields are listed. Finally, the graph-based knowledge distillation is summarized and prospectively discussed. We have also released related resources at https://github.com/liujing1023/Graph-based-Knowledge-Distillation.
Large vision Transformers (ViTs) driven by self-supervised pre-training mechanisms achieved unprecedented progress. Lightweight ViT models limited by the model capacity, however, benefit little from those pre-training mechanisms. Knowledge distillation defines a paradigm to transfer representations from large (teacher) models to small (student) ones. However, the conventional single-stage distillation easily gets stuck on task-specific transfer, failing to retain the task-agnostic knowledge crucial for model generalization. In this study, we propose generic-to-specific distillation (G2SD), to tap the potential of small ViT models under the supervision of large models pre-trained by masked autoencoders. In generic distillation, decoder of the small model is encouraged to align feature predictions with hidden representations of the large model, so that task-agnostic knowledge can be transferred. In specific distillation, predictions of the small model are constrained to be consistent with those of the large model, to transfer task-specific features which guarantee task performance. With G2SD, the vanilla ViT-Small model respectively achieves 98.7%, 98.1% and 99.3% the performance of its teacher (ViT-Base) for image classification, object detection, and semantic segmentation, setting a solid baseline for two-stage vision distillation. Code will be available at https://github.com/pengzhiliang/G2SD.
3D Semantic Scene Completion (SSC) can provide dense geometric and semantic scene representations, which can be applied in the field of autonomous driving and robotic systems. It is challenging to estimate the complete geometry and semantics of a scene solely from visual images, and accurate depth information is crucial for restoring 3D geometry. In this paper, we propose the first stereo SSC method named OccDepth, which fully exploits implicit depth information from stereo images (or RGBD images) to help the recovery of 3D geometric structures. The Stereo Soft Feature Assignment (Stereo-SFA) module is proposed to better fuse 3D depth-aware features by implicitly learning the correlation between stereo images. In particular, when the input are RGBD image, a virtual stereo images can be generated through original RGB image and depth map. Besides, the Occupancy Aware Depth (OAD) module is used to obtain geometry-aware 3D features by knowledge distillation using pre-trained depth models. In addition, a reformed TartanAir benchmark, named SemanticTartanAir, is provided in this paper for further testing our OccDepth method on SSC task. Compared with the state-of-the-art RGB-inferred SSC method, extensive experiments on SemanticKITTI show that our OccDepth method achieves superior performance with improving +4.82% mIoU, of which +2.49% mIoU comes from stereo images and +2.33% mIoU comes from our proposed depth-aware method. Our code and trained models are available at https://github.com/megvii-research/OccDepth.
The COVID-19 pandemic has caused substantial damage to global health. Even though three years have passed, the world continues to struggle with the virus. Concerns are growing about the impact of COVID-19 on the mental health of infected individuals, who are more likely to experience depression, which can have long-lasting consequences for both the affected individuals and the world. Detection and intervention at an early stage can reduce the risk of depression in COVID-19 patients. In this paper, we investigated the relationship between COVID-19 infection and depression through social media analysis. Firstly, we managed a dataset of COVID-19 patients that contains information about their social media activity both before and after infection. Secondly,We conducted an extensive analysis of this dataset to investigate the characteristic of COVID-19 patients with a higher risk of depression. Thirdly, we proposed a deep neural network for early prediction of depression risk. This model considers daily mood swings as a psychiatric signal and incorporates textual and emotional characteristics via knowledge distillation. Experimental results demonstrate that our proposed framework outperforms baselines in detecting depression risk, with an AUROC of 0.9317 and an AUPRC of 0.8116. Our model has the potential to enable public health organizations to initiate prompt intervention with high-risk patients
Named Entity Recognition (NER) models capable of Continual Learning (CL) are realistically valuable in areas where entity types continuously increase (e.g., personal assistants). Meanwhile the learning paradigm of NER advances to new patterns such as the span-based methods. However, its potential to CL has not been fully explored. In this paper, we propose SpanKL, a simple yet effective Span-based model with Knowledge distillation (KD) to preserve memories and multi-Label prediction to prevent conflicts in CL-NER. Unlike prior sequence labeling approaches, the inherently independent modeling in span and entity level with the designed coherent optimization on SpanKL promotes its learning at each incremental step and mitigates the forgetting. Experiments on synthetic CL datasets derived from OntoNotes and Few-NERD show that SpanKL significantly outperforms previous SoTA in many aspects, and obtains the smallest gap from CL to the upper bound revealing its high practiced value. The code is available at https://github.com/Qznan/SpanKL.
When deployed for risk-sensitive tasks, deep neural networks must be able to detect instances with labels from outside the distribution for which they were trained. In this paper we present a novel framework to benchmark the ability of image classifiers to detect class-out-of-distribution instances (i.e., instances whose true labels do not appear in the training distribution) at various levels of detection difficulty. We apply this technique to ImageNet, and benchmark 525 pretrained, publicly available, ImageNet-1k classifiers. The code for generating a benchmark for any ImageNet-1k classifier, along with the benchmarks prepared for the above-mentioned 525 models is available at https://github.com/mdabbah/COOD_benchmarking. The usefulness of the proposed framework and its advantage over alternative existing benchmarks is demonstrated by analyzing the results obtained for these models, which reveals numerous novel observations including: (1) knowledge distillation consistently improves class-out-of-distribution (C-OOD) detection performance; (2) a subset of ViTs performs better C-OOD detection than any other model; (3) the language--vision CLIP model achieves good zero-shot detection performance, with its best instance outperforming 96% of all other models evaluated; (4) accuracy and in-distribution ranking are positively correlated to C-OOD detection; and (5) we compare various confidence functions for C-OOD detection. Our companion paper, also published in ICLR 2023 (What Can We Learn From The Selective Prediction And Uncertainty Estimation Performance Of 523 Imagenet Classifiers), examines the uncertainty estimation performance (ranking, calibration, and selective prediction performance) of these classifiers in an in-distribution setting.
Pre-trained Transformer models such as BERT have shown great success in a wide range of applications, but at the cost of substantial increases in model complexity. Quantization-aware training (QAT) is a promising method to lower the implementation cost and energy consumption. However, aggressive quantization below 2-bit causes considerable accuracy degradation due to unstable convergence, especially when the downstream dataset is not abundant. This work proposes a proactive knowledge distillation method called Teacher Intervention (TI) for fast converging QAT of ultra-low precision pre-trained Transformers. TI intervenes layer-wise signal propagation with the intact signal from the teacher to remove the interference of propagated quantization errors, smoothing loss surface of QAT and expediting the convergence. Furthermore, we propose a gradual intervention mechanism to stabilize the recovery of subsections of Transformer layers from quantization. The proposed schemes enable fast convergence of QAT and improve the model accuracy regardless of the diverse characteristics of downstream fine-tuning tasks. We demonstrate that TI consistently achieves superior accuracy with significantly lower fine-tuning iterations on well-known Transformers of natural language processing as well as computer vision compared to the state-of-the-art QAT methods.
Scaled dot-product attention applies a softmax function on the scaled dot-product of queries and keys to calculate weights and then multiplies the weights and values. In this work, we study how to improve the learning of scaled dot-product attention to improve the accuracy of DETR. Our method is based on the following observations: using ground truth foreground-background mask (GT Fg-Bg Mask) as additional cues in the weights/values learning enables learning much better weights/values; with better weights/values, better values/weights can be learned. We propose a triple-attention module in which the first attention is a plain scaled dot-product attention, the second/third attention generates high-quality weights/values (with the assistance of GT Fg-Bg Mask) and shares the values/weights with the first attention to improve the quality of values/weights. The second and third attentions are removed during inference. We call our method knowledge-sharing DETR (KS-DETR), which is an extension of knowledge distillation (KD) in the way that the improved weights and values of the teachers (the second and third attentions) are directly shared, instead of mimicked, by the student (the first attention) to enable more efficient knowledge transfer from the teachers to the student. Experiments on various DETR-like methods show consistent improvements over the baseline methods on the MS COCO benchmark. Code is available at https://github.com/edocanonymous/KS-DETR.
Federated learning enables edge devices to train a global model collaboratively without exposing their data. Despite achieving outstanding advantages in computing efficiency and privacy protection, federated learning faces a significant challenge when dealing with non-IID data, i.e., data generated by clients that are typically not independent and identically distributed. In this paper, we tackle a new type of Non-IID data, called cluster-skewed non-IID, discovered in actual data sets. The cluster-skewed non-IID is a phenomenon in which clients can be grouped into clusters with similar data distributions. By performing an in-depth analysis of the behavior of a classification model's penultimate layer, we introduce a metric that quantifies the similarity between two clients' data distributions without violating their privacy. We then propose an aggregation scheme that guarantees equality between clusters. In addition, we offer a novel local training regularization based on the knowledge-distillation technique that reduces the overfitting problem at clients and dramatically boosts the training scheme's performance. We theoretically prove the superiority of the proposed aggregation over the benchmark FedAvg. Extensive experimental results on both standard public datasets and our in-house real-world dataset demonstrate that the proposed approach improves accuracy by up to 16% compared to the FedAvg algorithm.
The rise of mobile AI accelerators allows latency-sensitive applications to execute lightweight Deep Neural Networks (DNNs) on the client side. However, critical applications require powerful models that edge devices cannot host and must therefore offload requests, where the high-dimensional data will compete for limited bandwidth. This work proposes shifting away from focusing on executing shallow layers of partitioned DNNs. Instead, it advocates concentrating the local resources on variational compression optimized for machine interpretability. We introduce a novel framework for resource-conscious compression models and extensively evaluate our method in an environment reflecting the asymmetric resource distribution between edge devices and servers. Our method achieves 60% lower bitrate than a state-of-the-art SC method without decreasing accuracy and is up to 16x faster than offloading with existing codec standards.
Despite recommender systems play a key role in network content platforms, mining the user's interests is still a significant challenge. Existing works predict the user interest by utilizing user behaviors, i.e., clicks, views, etc., but current solutions are ineffective when users perform unsettled activities. The latter ones involve new users, which have few activities of any kind, and sparse users who have low-frequency behaviors. We uniformly describe both these user-types as "cold users", which are very common but often neglected in network content platforms. To address this issue, we enhance the representation of the user interest by combining his social interest, e.g., friendship, following bloggers, interest groups, etc., with the activity behaviors. Thus, in this work, we present a novel algorithm entitled SocialNet, which adopts a two-stage method to progressively extract the coarse-grained and fine-grained social interest. Our technique then concatenates SocialNet's output with the original user representation to get the final user representation that combines behavior interests and social interests. Offline experiments on Tencent video's recommender system demonstrate the superiority over the baseline behavior-based model. The online experiment also shows a significant performance improvement in clicks and view time in the real-world recommendation system. The source code is available at https://github.com/Social4Rec/SocialNet.
Currently, one of the major challenges in deep learning-based video frame interpolation (VFI) is the large model sizes and high computational complexity associated with many high performance VFI approaches. In this paper, we present a distillation-based two-stage workflow for obtaining compressed VFI models which perform competitively to the state of the arts, at a greatly reduced model size and complexity. Specifically, an optimisation-based network pruning method is first applied to a recently proposed frame interpolation model, ST-MFNet, which outperforms many other VFI methods but suffers from large model size. The resulting new network architecture achieves a 91% reduction in parameters and 35% increase in speed. Secondly, the performance of the new network is further enhanced through a teacher-student knowledge distillation training process using a Laplacian distillation loss. The final low complexity model, ST-MFNet Mini, achieves a comparable performance to most existing high-complex VFI methods, only outperformed by the original ST-MFNet. Our source code is available at https://github.com/crispianm/ST-MFNet-Mini
To imitate the ability of keeping learning of human, continual learning which can learn from a never-ending data stream has attracted more interests recently. In all settings, the online class incremental learning (OCIL), where incoming samples from data stream can be used only once, is more challenging and can be encountered more frequently in real world. Actually, all continual learning models face a stability-plasticity dilemma, where the stability means the ability to preserve old knowledge while the plasticity denotes the ability to incorporate new knowledge. Although replay-based methods have shown exceptional promise, most of them concentrate on the strategy for updating and retrieving memory to keep stability at the expense of plasticity. To strike a preferable trade-off between stability and plasticity, we propose an Adaptive Focus Shifting algorithm (AFS), which dynamically adjusts focus to ambiguous samples and non-target logits in model learning. Through a deep analysis of the task-recency bias caused by class imbalance, we propose a revised focal loss to mainly keep stability. \Rt{By utilizing a new weight function, the revised focal loss will pay more attention to current ambiguous samples, which are the potentially valuable samples to make model progress quickly.} To promote plasticity, we introduce a virtual knowledge distillation. By designing a virtual teacher, it assigns more attention to non-target classes, which can surmount overconfidence and encourage model to focus on inter-class information. Extensive experiments on three popular datasets for OCIL have shown the effectiveness of AFS. The code will be available at \url{https://github.com/czjghost/AFS}.
Deep learning has contributed greatly to many successes in artificial intelligence in recent years. Today, it is possible to train models that have thousands of layers and hundreds of billions of parameters. Large-scale deep models have achieved great success, but the enormous computational complexity and gigantic storage requirements make it extremely difficult to implement them in real-time applications. On the other hand, the size of the dataset is still a real problem in many domains. Data are often missing, too expensive, or impossible to obtain for other reasons. Ensemble learning is partially a solution to the problem of small datasets and overfitting. However, ensemble learning in its basic version is associated with a linear increase in computational complexity. We analyzed the impact of the ensemble decision-fusion mechanism and checked various methods of sharing the decisions including voting algorithms. We used the modified knowledge distillation framework as a decision-fusion mechanism which allows in addition compressing of the entire ensemble model into a weight space of a single model. We showed that knowledge distillation can aggregate knowledge from multiple teachers in only one student model and, with the same computational complexity, obtain a better-performing model compared to a model trained in the standard manner. We have developed our own method for mimicking the responses of all teachers at the same time, simultaneously. We tested these solutions on several benchmark datasets. In the end, we presented a wide application use of the efficient multi-teacher knowledge distillation framework. In the first example, we used knowledge distillation to develop models that could automate corrosion detection on aircraft fuselage. The second example describes detection of smoke on observation cameras in order to counteract wildfires in forests.
Personalized Federated Learning (pFL) has emerged as a promising solution to tackle data heterogeneity across clients in FL. However, existing pFL methods either (1) introduce high communication and computation costs or (2) overfit to local data, which can be limited in scope, and are vulnerable to evolved test samples with natural shifts. In this paper, we propose PerAda, a parameter-efficient pFL framework that reduces communication and computational costs and exhibits superior generalization performance, especially under test-time distribution shifts. PerAda reduces the costs by leveraging the power of pretrained models and only updates and communicates a small number of additional parameters from adapters. PerAda has good generalization since it regularizes each client's personalized adapter with a global adapter, while the global adapter uses knowledge distillation to aggregate generalized information from all clients. Theoretically, we provide generalization bounds to explain why PerAda improves generalization, and we prove its convergence to stationary points under non-convex settings. Empirically, PerAda demonstrates competitive personalized performance (+4.85% on CheXpert) and enables better out-of-distribution generalization (+5.23% on CIFAR-10-C) on different datasets across natural and medical domains compared with baselines, while only updating 12.6% of parameters per model based on the adapter. Our code is available at https://github.com/NVlabs/PerAda.
Knowledge distillation is an effective method for model compression. However, it is still a challenging topic to apply knowledge distillation to detection tasks. There are two key points resulting in poor distillation performance for detection tasks. One is the serious imbalance between foreground and background features, another one is that small object lacks enough feature representation. To solve the above issues, we propose a new distillation method named dual relation knowledge distillation (DRKD), including pixel-wise relation distillation and instance-wise relation distillation. The pixel-wise relation distillation embeds pixel-wise features in the graph space and applies graph convolution to capture the global pixel relation. By distilling the global pixel relation, the student detector can learn the relation between foreground and background features, and avoid the difficulty of distilling features directly for the feature imbalance issue. Besides, we find that instance-wise relation supplements valuable knowledge beyond independent features for small objects. Thus, the instance-wise relation distillation is designed, which calculates the similarity of different instances to obtain a relation matrix. More importantly, a relation filter module is designed to highlight valuable instance relations. The proposed dual relation knowledge distillation is general and can be easily applied for both one-stage and two-stage detectors. Our method achieves state-of-the-art performance, which improves Faster R-CNN based on ResNet50 from 38.4% to 41.6% mAP and improves RetinaNet based on ResNet50 from 37.4% to 40.3% mAP on COCO 2017.