PyTorchKR Papers

최신 논문

총 4,526편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2309.02843공식 코드 · Papers with Code

Knowledge Distillation Layer that Lets the Student Decide

Typical technique in knowledge distillation (KD) is regularizing the learning of a limited capacity model (student) by pushing its responses to match a powerful model's (teacher). Albeit useful especially in the penultimate layer and beyond, its action on student's feature transform is rather implicit, limiting its practice in the intermediate layers. To explicitly embed the teacher's knowledge in feature transform, we propose a learnable KD layer for the student which improves KD with two distinct abilities: i) learning how to leverage the teacher's knowledge, enabling to discard nuisance information, and ii) feeding forward the transferred knowledge deeper. Thus, the student enjoys the teacher's knowledge during the inference besides training. Formally, we repurpose 1x1-BN-ReLU-1x1 convolution block to assign a semantic vector to each local region according to the template (supervised by the teacher) that the corresponding region of the student matches. To facilitate template learning in the intermediate layers, we propose a novel form of supervision based on the teacher's decisions. Through rigorous experimentation, we demonstrate the effectiveness of our approach on 3 popular classification benchmarks. Code is available at: https://github.com/adagorgun/letKD-framework

arXiv:2309.01682

Prior Knowledge Guided Network for Video Anomaly Detection

Video Anomaly Detection (VAD) involves detecting anomalous events in videos, presenting a significant and intricate task within intelligent video surveillance. Existing studies often concentrate solely on features acquired from limited normal data, disregarding the latent prior knowledge present in extensive natural image datasets. To address this constraint, we propose a Prior Knowledge Guided Network(PKG-Net) for the VAD task. First, an auto-encoder network is incorporated into a teacher-student architecture to learn two designated proxy tasks: future frame prediction and teacher network imitation, which can provide better generalization ability on unknown samples. Second, knowledge distillation on proper feature blocks is also proposed to increase the multi-scale detection ability of the model. In addition, prediction error and teacher-student feature inconsistency are combined to evaluate anomaly scores of inference samples more comprehensively. Experimental results on three public benchmarks validate the effectiveness and accuracy of our method, which surpasses recent state-of-the-arts.

arXiv:2309.01612

On the Query Strategies for Efficient Online Active Distillation

Deep Learning (DL) requires lots of time and data, resulting in high computational demands. Recently, researchers employ Active Learning (AL) and online distillation to enhance training efficiency and real-time model adaptation. This paper evaluates a set of query strategies to achieve the best training results. It focuses on Human Pose Estimation (HPE) applications, assessing the impact of selected frames during training using two approaches: a classical offline method and a online evaluation through a continual learning approach employing knowledge distillation, on a popular state-of-the-art HPE dataset. The paper demonstrates the possibility of enabling training at the edge lightweight models, adapting them effectively to new contexts in real-time.

arXiv:2309.02719

DMKD: Improving Feature-based Knowledge Distillation for Object Detection Via Dual Masking Augmentation

Recent mainstream masked distillation methods function by reconstructing selectively masked areas of a student network from the feature map of its teacher counterpart. In these methods, the masked regions need to be properly selected, such that reconstructed features encode sufficient discrimination and representation capability like the teacher feature. However, previous masked distillation methods only focus on spatial masking, making the resulting masked areas biased towards spatial importance without encoding informative channel clues. In this study, we devise a Dual Masked Knowledge Distillation (DMKD) framework which can capture both spatially important and channel-wise informative clues for comprehensive masked feature reconstruction. More specifically, we employ dual attention mechanism for guiding the respective masking branches, leading to reconstructed feature encoding dual significance. Furthermore, fusing the reconstructed features is achieved by self-adjustable weighting strategy for effective feature distillation. Our experiments on object detection task demonstrate that the student networks achieve performance gains of 4.1% and 4.3% with the help of our method when RetinaNet and Cascade Mask R-CNN are respectively used as the teacher networks, while outperforming the other state-of-the-art distillation methods.

arXiv:2309.04195공식 코드 · Papers with Code

Towards Mitigating Architecture Overfitting on Distilled Datasets

Dataset distillation methods have demonstrated remarkable performance for neural networks trained with very limited training data. However, a significant challenge arises in the form of \textit{architecture overfitting}: the distilled training dataset synthesized by a specific network architecture (i.e., training network) generates poor performance when trained by other network architectures (i.e., test networks), especially when the test networks have a larger capacity than the training network. This paper introduces a series of approaches to mitigate this issue. Among them, DropPath renders the large model to be an implicit ensemble of its sub-networks, and knowledge distillation ensures each sub-network acts similarly to the small but well-performing teacher network. These methods, characterized by their smoothing effects, significantly mitigate architecture overfitting. We conduct extensive experiments to demonstrate the effectiveness and generality of our methods. Particularly, across various scenarios involving different tasks and different sizes of distilled data, our approaches significantly mitigate architecture overfitting. Furthermore, our approaches achieve comparable or even superior performance when the test network is larger than the training network.

arXiv:2309.04062

3D Denoisers are Good 2D Teachers: Molecular Pretraining via Denoising and Cross-Modal Distillation

Pretraining molecular representations from large unlabeled data is essential for molecular property prediction due to the high cost of obtaining ground-truth labels. While there exist various 2D graph-based molecular pretraining approaches, these methods struggle to show statistically significant gains in predictive performance. Recent work have thus instead proposed 3D conformer-based pretraining under the task of denoising, which led to promising results. During downstream finetuning, however, models trained with 3D conformers require accurate atom-coordinates of previously unseen molecules, which are computationally expensive to acquire at scale. In light of this limitation, we propose D&D, a self-supervised molecular representation learning framework that pretrains a 2D graph encoder by distilling representations from a 3D denoiser. With denoising followed by cross-modal knowledge distillation, our approach enjoys use of knowledge obtained from denoising as well as painless application to downstream tasks with no access to accurate conformers. Experiments on real-world molecular property prediction datasets show that the graph encoder trained via D&D can infer 3D information based on the 2D graph and shows superior performance and label-efficiency against other baselines.

arXiv:2309.02887

A deep Natural Language Inference predictor without language-specific training data

In this paper we present a technique of NLP to tackle the problem of inference relation (NLI) between pairs of sentences in a target language of choice without a language-specific training dataset. We exploit a generic translation dataset, manually translated, along with two instances of the same pre-trained model - the first to generate sentence embeddings for the source language, and the second fine-tuned over the target language to mimic the first. This technique is known as Knowledge Distillation. The model has been evaluated over machine translated Stanford NLI test dataset, machine translated Multi-Genre NLI test dataset, and manually translated RTE3-ITA test dataset. We also test the proposed architecture over different tasks to empirically demonstrate the generality of the NLI task. The model has been evaluated over the native Italian ABSITA dataset, on the tasks of Sentiment Analysis, Aspect-Based Sentiment Analysis, and Topic Recognition. We emphasise the generality and exploitability of the Knowledge Distillation technique that outperforms other methodologies based on machine translation, even though the former was not directly trained on the data it was tested over.

arXiv:2309.02870공식 코드 · Papers with Code

Rethinking Momentum Knowledge Distillation in Online Continual Learning

Online Continual Learning (OCL) addresses the problem of training neural networks on a continuous data stream where multiple classification tasks emerge in sequence. In contrast to offline Continual Learning, data can be seen only once in OCL, which is a very severe constraint. In this context, replay-based strategies have achieved impressive results and most state-of-the-art approaches heavily depend on them. While Knowledge Distillation (KD) has been extensively used in offline Continual Learning, it remains under-exploited in OCL, despite its high potential. In this paper, we analyze the challenges in applying KD to OCL and give empirical justifications. We introduce a direct yet effective methodology for applying Momentum Knowledge Distillation (MKD) to many flagship OCL methods and demonstrate its capabilities to enhance existing approaches. In addition to improving existing state-of-the-art accuracy by more than $10\%$ points on ImageNet100, we shed light on MKD internal mechanics and impacts during training in OCL. We argue that similar to replay, MKD should be considered a central component of OCL. The code is available at \url{https://github.com/Nicolas1203/mkd_ocl}.

arXiv:2309.03659

Towards Comparable Knowledge Distillation in Semantic Image Segmentation

Knowledge Distillation (KD) is one proposed solution to large model sizes and slow inference speed in semantic segmentation. In our research we identify 25 proposed distillation loss terms from 14 publications in the last 4 years. Unfortunately, a comparison of terms based on published results is often impossible, because of differences in training configurations. A good illustration of this problem is the comparison of two publications from 2022. Using the same models and dataset, Structural and Statistical Texture Distillation (SSTKD) reports an increase of student mIoU of 4.54 and a final performance of 29.19, while Adaptive Perspective Distillation (APD) only improves student performance by 2.06 percentage points, but achieves a final performance of 39.25. The reason for such extreme differences is often a suboptimal choice of hyperparameters and a resulting underperformance of the student model used as reference point. In our work, we reveal problems of insufficient hyperparameter tuning by showing that distillation improvements of two widely accepted frameworks, SKD and IFVD, vanish when hyperparameters are optimized sufficiently. To improve comparability of future research in the field, we establish a solid baseline for three datasets and two student models and provide extensive information on hyperparameter tuning. We find that only two out of eight techniques can compete with our simple baseline on the ADE20K dataset.

arXiv:2309.02031

A survey on efficient vision transformers: algorithms, techniques, and performance benchmarking

Vision Transformer (ViT) architectures are becoming increasingly popular and widely employed to tackle computer vision applications. Their main feature is the capacity to extract global information through the self-attention mechanism, outperforming earlier convolutional neural networks. However, ViT deployment and performance have grown steadily with their size, number of trainable parameters, and operations. Furthermore, self-attention's computational and memory cost quadratically increases with the image resolution. Generally speaking, it is challenging to employ these architectures in real-world applications due to many hardware and environmental restrictions, such as processing and computational capabilities. Therefore, this survey investigates the most efficient methodologies to ensure sub-optimal estimation performances. More in detail, four efficient categories will be analyzed: compact architecture, pruning, knowledge distillation, and quantization strategies. Moreover, a new metric called Efficient Error Rate has been introduced in order to normalize and compare models' features that affect hardware devices at inference time, such as the number of parameters, bits, FLOPs, and model size. Summarizing, this paper firstly mathematically defines the strategies used to make Vision Transformer efficient, describes and discusses state-of-the-art methodologies, and analyzes their performances over different application scenarios. Toward the end of this paper, we also discuss open challenges and promising research directions.

arXiv:2309.01270공식 코드 · Papers with Code

COMEDIAN: Self-Supervised Learning and Knowledge Distillation for Action Spotting using Transformers

We present COMEDIAN, a novel pipeline to initialize spatiotemporal transformers for action spotting, which involves self-supervised learning and knowledge distillation. Action spotting is a timestamp-level temporal action detection task. Our pipeline consists of three steps, with two initialization stages. First, we perform self-supervised initialization of a spatial transformer using short videos as input. Additionally, we initialize a temporal transformer that enhances the spatial transformer's outputs with global context through knowledge distillation from a pre-computed feature bank aligned with each short video segment. In the final step, we fine-tune the transformers to the action spotting task. The experiments, conducted on the SoccerNet-v2 dataset, demonstrate state-of-the-art performance and validate the effectiveness of COMEDIAN's pretraining paradigm. Our results highlight several advantages of our pretraining pipeline, including improved performance and faster convergence compared to non-pretrained models.

arXiv:2308.16415

Knowledge Distillation from Non-streaming to Streaming ASR Encoder using Auxiliary Non-streaming Layer

Streaming automatic speech recognition (ASR) models are restricted from accessing future context, which results in worse performance compared to the non-streaming models. To improve the performance of streaming ASR, knowledge distillation (KD) from the non-streaming to streaming model has been studied, mainly focusing on aligning the output token probabilities. In this paper, we propose a layer-to-layer KD from the teacher encoder to the student encoder. To ensure that features are extracted using the same context, we insert auxiliary non-streaming branches to the student and perform KD from the non-streaming teacher layer to the non-streaming auxiliary layer. We design a special KD loss that leverages the autoregressive predictive coding (APC) mechanism to encourage the streaming model to predict unseen future contexts. Experimental results show that the proposed method can significantly reduce the word error rate compared to previous token probability distillation methods.

arXiv:2308.16454

Adversarial Finetuning with Latent Representation Constraint to Mitigate Accuracy-Robustness Tradeoff

This paper addresses the tradeoff between standard accuracy on clean examples and robustness against adversarial examples in deep neural networks (DNNs). Although adversarial training (AT) improves robustness, it degrades the standard accuracy, thus yielding the tradeoff. To mitigate this tradeoff, we propose a novel AT method called ARREST, which comprises three components: (i) adversarial finetuning (AFT), (ii) representation-guided knowledge distillation (RGKD), and (iii) noisy replay (NR). AFT trains a DNN on adversarial examples by initializing its parameters with a DNN that is standardly pretrained on clean examples. RGKD and NR respectively entail a regularization term and an algorithm to preserve latent representations of clean examples during AFT. RGKD penalizes the distance between the representations of the standardly pretrained and AFT DNNs. NR switches input adversarial examples to nonadversarial ones when the representation changes significantly during AFT. By combining these components, ARREST achieves both high standard accuracy and robustness. Experimental results demonstrate that ARREST mitigates the tradeoff more effectively than previous AT-based methods do.

arXiv:2308.16561공식 코드 · Papers with Code

MoMA: Momentum Contrastive Learning with Multi-head Attention-based Knowledge Distillation for Histopathology Image Analysis

There is no doubt that advanced artificial intelligence models and high quality data are the keys to success in developing computational pathology tools. Although the overall volume of pathology data keeps increasing, a lack of quality data is a common issue when it comes to a specific task due to several reasons including privacy and ethical issues with patient data. In this work, we propose to exploit knowledge distillation, i.e., utilize the existing model to learn a new, target model, to overcome such issues in computational pathology. Specifically, we employ a student-teacher framework to learn a target model from a pre-trained, teacher model without direct access to source data and distill relevant knowledge via momentum contrastive learning with multi-head attention mechanism, which provides consistent and context-aware feature representations. This enables the target model to assimilate informative representations of the teacher model while seamlessly adapting to the unique nuances of the target data. The proposed method is rigorously evaluated across different scenarios where the teacher model was trained on the same, relevant, and irrelevant classification tasks with the target model. Experimental results demonstrate the accuracy and robustness of our approach in transferring knowledge to different domains and tasks, outperforming other related methods. Moreover, the results provide a guideline on the learning strategy for different types of tasks and scenarios in computational pathology. Code is available at: \url{https://github.com/trinhvg/MoMA}.

arXiv:2308.16609

Towards Long-Tailed Recognition for Graph Classification via Collaborative Experts

Graph classification, aiming at learning the graph-level representations for effective class assignments, has received outstanding achievements, which heavily relies on high-quality datasets that have balanced class distribution. In fact, most real-world graph data naturally presents a long-tailed form, where the head classes occupy much more samples than the tail classes, it thus is essential to study the graph-level classification over long-tailed data while still remaining largely unexplored. However, most existing long-tailed learning methods in visions fail to jointly optimize the representation learning and classifier training, as well as neglect the mining of the hard-to-classify classes. Directly applying existing methods to graphs may lead to sub-optimal performance, since the model trained on graphs would be more sensitive to the long-tailed distribution due to the complex topological characteristics. Hence, in this paper, we propose a novel long-tailed graph-level classification framework via Collaborative Multi-expert Learning (CoMe) to tackle the problem. To equilibrate the contributions of head and tail classes, we first develop balanced contrastive learning from the view of representation learning, and then design an individual-expert classifier training based on hard class mining. In addition, we execute gated fusion and disentangled knowledge distillation among the multiple experts to promote the collaboration in a multi-expert framework. Comprehensive experiments are performed on seven widely-used benchmark datasets to demonstrate the superiority of our method CoMe over state-of-the-art baselines.

arXiv:2308.15846

Exploring Multi-Modal Contextual Knowledge for Open-Vocabulary Object Detection

In this paper, we for the first time explore helpful multi-modal contextual knowledge to understand novel categories for open-vocabulary object detection (OVD). The multi-modal contextual knowledge stands for the joint relationship across regions and words. However, it is challenging to incorporate such multi-modal contextual knowledge into OVD. The reason is that previous detection frameworks fail to jointly model multi-modal contextual knowledge, as object detectors only support vision inputs and no caption description is provided at test time. To this end, we propose a multi-modal contextual knowledge distillation framework, MMC-Det, to transfer the learned contextual knowledge from a teacher fusion transformer with diverse multi-modal masked language modeling (D-MLM) to a student detector. The diverse multi-modal masked language modeling is realized by an object divergence constraint upon traditional multi-modal masked language modeling (MLM), in order to extract fine-grained region-level visual contexts, which are vital to object detection. Extensive experiments performed upon various detection datasets show the effectiveness of our multi-modal context learning strategy, where our approach well outperforms the recent state-of-the-art methods.

arXiv:2308.14852공식 코드 · Papers with Code

SynthDistill: Face Recognition with Knowledge Distillation from Synthetic Data

State-of-the-art face recognition networks are often computationally expensive and cannot be used for mobile applications. Training lightweight face recognition models also requires large identity-labeled datasets. Meanwhile, there are privacy and ethical concerns with collecting and using large face recognition datasets. While generating synthetic datasets for training face recognition models is an alternative option, it is challenging to generate synthetic data with sufficient intra-class variations. In addition, there is still a considerable gap between the performance of models trained on real and synthetic data. In this paper, we propose a new framework (named SynthDistill) to train lightweight face recognition models by distilling the knowledge of a pretrained teacher face recognition model using synthetic data. We use a pretrained face generator network to generate synthetic face images and use the synthesized images to learn a lightweight student network. We use synthetic face images without identity labels, mitigating the problems in the intra-class variation generation of synthetic datasets. Instead, we propose a novel dynamic sampling strategy from the intermediate latent space of the face generator network to include new variations of the challenging images while further exploring new face images in the training batch. The results on five different face recognition datasets demonstrate the superiority of our lightweight model compared to models trained on previous synthetic datasets, achieving a verification accuracy of 99.52% on the LFW dataset with a lightweight network. The results also show that our proposed framework significantly reduces the gap between training with real and synthetic data. The source code for replicating the experiments is publicly released.

arXiv:2308.15122공식 코드 · Papers with Code

SpikeBERT: A Language Spikformer Learned from BERT with Knowledge Distillation

Spiking neural networks (SNNs) offer a promising avenue to implement deep neural networks in a more energy-efficient way. However, the network architectures of existing SNNs for language tasks are still simplistic and relatively shallow, and deep architectures have not been fully explored, resulting in a significant performance gap compared to mainstream transformer-based networks such as BERT. To this end, we improve a recently-proposed spiking Transformer (i.e., Spikformer) to make it possible to process language tasks and propose a two-stage knowledge distillation method for training it, which combines pre-training by distilling knowledge from BERT with a large collection of unlabelled texts and fine-tuning with task-specific instances via knowledge distillation again from the BERT fine-tuned on the same training examples. Through extensive experimentation, we show that the models trained with our method, named SpikeBERT, outperform state-of-the-art SNNs and even achieve comparable results to BERTs on text classification tasks for both English and Chinese with much less energy consumption. Our code is available at https://github.com/Lvchangze/SpikeBERT.

arXiv:2308.13662

REFT: Resource-Efficient Federated Training Framework for Heterogeneous and Resource-Constrained Environments

Federated Learning (FL) plays a critical role in distributed systems. In these systems, data privacy and confidentiality hold paramount importance, particularly within edge-based data processing systems such as IoT devices deployed in smart homes. FL emerges as a privacy-enforcing sub-domain of machine learning that enables model training on client devices, eliminating the necessity to share private data with a central server. While existing research has predominantly addressed challenges pertaining to data heterogeneity, there remains a current gap in addressing issues such as varying device capabilities and efficient communication. These unaddressed issues raise a number of implications in resource-constrained environments. In particular, the practical implementation of FL-based IoT or edge systems is extremely inefficient. In this paper, we propose "Resource-Efficient Federated Training Framework for Heterogeneous and Resource-Constrained Environments (REFT)," a novel approach specifically devised to address these challenges in resource-limited devices. Our proposed method uses Variable Pruning to optimize resource utilization by adapting pruning strategies to the computational capabilities of each client. Furthermore, our proposed REFT technique employs knowledge distillation to minimize the need for continuous bidirectional client-server communication. This achieves a significant reduction in communication bandwidth, thereby enhancing the overall resource efficiency. We conduct experiments for an image classification task, and the results demonstrate the effectiveness of our approach in resource-limited settings. Our technique not only preserves data privacy and performance standards but also accommodates heterogeneous model architectures, facilitating the participation of a broader array of diverse client devices in the training process, all while consuming minimal bandwidth.

arXiv:2308.13798공식 코드 · Papers with Code

DM-VTON: Distilled Mobile Real-time Virtual Try-On

The fashion e-commerce industry has witnessed significant growth in recent years, prompting exploring image-based virtual try-on techniques to incorporate Augmented Reality (AR) experiences into online shopping platforms. However, existing research has primarily overlooked a crucial aspect - the runtime of the underlying machine-learning model. While existing methods prioritize enhancing output quality, they often disregard the execution time, which restricts their applications on a limited range of devices. To address this gap, we propose Distilled Mobile Real-time Virtual Try-On (DM-VTON), a novel virtual try-on framework designed to achieve simplicity and efficiency. Our approach is based on a knowledge distillation scheme that leverages a strong Teacher network as supervision to guide a Student network without relying on human parsing. Notably, we introduce an efficient Mobile Generative Module within the Student network, significantly reducing the runtime while ensuring high-quality output. Additionally, we propose Virtual Try-on-guided Pose for Data Synthesis to address the limited pose variation observed in training images. Experimental results show that the proposed method can achieve 40 frames per second on a single Nvidia Tesla T4 GPU and only take up 37 MB of memory while producing almost the same output quality as other state-of-the-art methods. DM-VTON stands poised to facilitate the advancement of real-time AR applications, in addition to the generation of lifelike attired human figures tailored for diverse specialized training tasks. https://sites.google.com/view/ltnghia/research/DMVTON