PyTorchKR Papers

최신 논문

총 4,526편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2401.11824공식 코드 · Papers with Code

Rethinking Centered Kernel Alignment in Knowledge Distillation

Knowledge distillation has emerged as a highly effective method for bridging the representation discrepancy between large-scale models and lightweight models. Prevalent approaches involve leveraging appropriate metrics to minimize the divergence or distance between the knowledge extracted from the teacher model and the knowledge learned by the student model. Centered Kernel Alignment (CKA) is widely used to measure representation similarity and has been applied in several knowledge distillation methods. However, these methods are complex and fail to uncover the essence of CKA, thus not answering the question of how to use CKA to achieve simple and effective distillation properly. This paper first provides a theoretical perspective to illustrate the effectiveness of CKA, which decouples CKA to the upper bound of Maximum Mean Discrepancy~(MMD) and a constant term. Drawing from this, we propose a novel Relation-Centered Kernel Alignment~(RCKA) framework, which practically establishes a connection between CKA and MMD. Furthermore, we dynamically customize the application of CKA based on the characteristics of each task, with less computational source yet comparable performance than the previous methods. The extensive experiments on the CIFAR-100, ImageNet-1k, and MS-COCO demonstrate that our method achieves state-of-the-art performance on almost all teacher-student pairs for image classification and object detection, validating the effectiveness of our approaches. Our code is available in https://github.com/Klayand/PCKA

arXiv:2401.12014

Robustness to distribution shifts of compressed networks for edge devices

It is necessary to develop efficient DNNs deployed on edge devices with limited computation resources. However, the compressed networks often execute new tasks in the target domain, which is different from the source domain where the original network is trained. It is important to investigate the robustness of compressed networks in two types of data distribution shifts: domain shifts and adversarial perturbations. In this study, we discover that compressed models are less robust to distribution shifts than their original networks. Interestingly, larger networks are more vulnerable to losing robustness than smaller ones, even when they are compressed to a similar size as the smaller networks. Furthermore, compact networks obtained by knowledge distillation are much more robust to distribution shifts than pruned networks. Finally, post-training quantization is a reliable method for achieving significant robustness to distribution shifts, and it outperforms both pruned and distilled models in terms of robustness.

arXiv:2401.11798

Knowledge Distillation on Spatial-Temporal Graph Convolutional Network for Traffic Prediction

Efficient real-time traffic prediction is crucial for reducing transportation time. To predict traffic conditions, we employ a spatio-temporal graph neural network (ST-GNN) to model our real-time traffic data as temporal graphs. Despite its capabilities, it often encounters challenges in delivering efficient real-time predictions for real-world traffic data. Recognizing the significance of timely prediction due to the dynamic nature of real-time data, we employ knowledge distillation (KD) as a solution to enhance the execution time of ST-GNNs for traffic prediction. In this paper, We introduce a cost function designed to train a network with fewer parameters (the student) using distilled data from a complex network (the teacher) while maintaining its accuracy close to that of the teacher. We use knowledge distillation, incorporating spatial-temporal correlations from the teacher network to enable the student to learn the complex patterns perceived by the teacher. However, a challenge arises in determining the student network architecture rather than considering it inadvertently. To address this challenge, we propose an algorithm that utilizes the cost function to calculate pruning scores, addressing small network architecture search issues, and jointly fine-tunes the network resulting from each pruning stage using KD. Ultimately, we evaluate our proposed ideas on two real-world datasets, PeMSD7 and PeMSD8. The results indicate that our method can maintain the student's accuracy close to that of the teacher, even with the retention of only 3% of network parameters.

arXiv:2401.11633

Zoom-shot: Fast and Efficient Unsupervised Zero-Shot Transfer of CLIP to Vision Encoders with Multimodal Loss

The fusion of vision and language has brought about a transformative shift in computer vision through the emergence of Vision-Language Models (VLMs). However, the resource-intensive nature of existing VLMs poses a significant challenge. We need an accessible method for developing the next generation of VLMs. To address this issue, we propose Zoom-shot, a novel method for transferring the zero-shot capabilities of CLIP to any pre-trained vision encoder. We do this by exploiting the multimodal information (i.e. text and image) present in the CLIP latent space through the use of specifically designed multimodal loss functions. These loss functions are (1) cycle-consistency loss and (2) our novel prompt-guided knowledge distillation loss (PG-KD). PG-KD combines the concept of knowledge distillation with CLIP's zero-shot classification, to capture the interactions between text and image features. With our multimodal losses, we train a $\textbf{linear mapping}$ between the CLIP latent space and the latent space of a pre-trained vision encoder, for only a $\textbf{single epoch}$. Furthermore, Zoom-shot is entirely unsupervised and is trained using $\textbf{unpaired}$ data. We test the zero-shot capabilities of a range of vision encoders augmented as new VLMs, on coarse and fine-grained classification datasets, outperforming the previous state-of-the-art in this problem domain. In our ablations, we find Zoom-shot allows for a trade-off between data and compute during training; and our state-of-the-art results can be obtained by reducing training from 20% to 1% of the ImageNet training data with 20 epochs. All code and models are available on GitHub.

arXiv:2401.10752공식 코드 · Papers with Code

HiCD: Change Detection in Quality-Varied Images via Hierarchical Correlation Distillation

Advanced change detection techniques primarily target image pairs of equal and high quality. However, variations in imaging conditions and platforms frequently lead to image pairs with distinct qualities: one image being high-quality, while the other being low-quality. These disparities in image quality present significant challenges for understanding image pairs semantically and extracting change features, ultimately resulting in a notable decline in performance. To tackle this challenge, we introduce an innovative training strategy grounded in knowledge distillation. The core idea revolves around leveraging task knowledge acquired from high-quality image pairs to guide the model's learning process when dealing with image pairs that exhibit differences in quality. Additionally, we develop a hierarchical correlation distillation approach (involving self-correlation, cross-correlation, and global correlation). This approach compels the student model to replicate the correlations inherent in the teacher model, rather than focusing solely on individual features. This ensures effective knowledge transfer while maintaining the student model's training flexibility.

arXiv:2401.10484

Enhancing Scalability in Recommender Systems through Lottery Ticket Hypothesis and Knowledge Distillation-based Neural Network Pruning

This study introduces an innovative approach aimed at the efficient pruning of neural networks, with a particular focus on their deployment on edge devices. Our method involves the integration of the Lottery Ticket Hypothesis (LTH) with the Knowledge Distillation (KD) framework, resulting in the formulation of three distinct pruning models. These models have been developed to address scalability issue in recommender systems, whereby the complexities of deep learning models have hindered their practical deployment. With judicious application of the pruning techniques, we effectively curtail the power consumption and model dimensions without compromising on accuracy. Empirical evaluation has been performed using two real world datasets from diverse domains against two baselines. Gratifyingly, our approaches yielded a GPU computation-power reduction of up to 66.67%. Notably, our study contributes to the field of recommendation system by pioneering the application of LTH and KD.

arXiv:2401.10446공식 코드 · Papers with Code

Large Language Models are Efficient Learners of Noise-Robust Speech Recognition

Recent advances in large language models (LLMs) have promoted generative error correction (GER) for automatic speech recognition (ASR), which leverages the rich linguistic knowledge and powerful reasoning ability of LLMs to improve recognition results. The latest work proposes a GER benchmark with HyPoradise dataset to learn the mapping from ASR N-best hypotheses to ground-truth transcription by efficient LLM finetuning, which shows great effectiveness but lacks specificity on noise-robust ASR. In this work, we extend the benchmark to noisy conditions and investigate if we can teach LLMs to perform denoising for GER just like what robust ASR do}, where one solution is introducing noise information as a conditioner into LLM. However, directly incorporating noise embeddings from audio encoder could harm the LLM tuning due to cross-modality gap. To this end, we propose to extract a language-space noise embedding from the N-best list to represent the noise conditions of source speech, which can promote the denoising process in GER. Furthermore, in order to enhance its representation ability of audio noise, we design a knowledge distillation (KD) approach via mutual information estimation to distill the real noise information in audio embeddings to our language embedding. Experiments on various latest LLMs demonstrate our approach achieves a new breakthrough with up to 53.9% correction improvement in terms of word error rate while with limited training data. Analysis shows that our language-space noise embedding can well represent the noise conditions of source speech, under which off-the-shelf LLMs show strong ability of language-space denoising.

arXiv:2401.09442

Object Attribute Matters in Visual Question Answering

Visual question answering is a multimodal task that requires the joint comprehension of visual and textual information. However, integrating visual and textual semantics solely through attention layers is insufficient to comprehensively understand and align information from both modalities. Intuitively, object attributes can naturally serve as a bridge to unify them, which has been overlooked in previous research. In this paper, we propose a novel VQA approach from the perspective of utilizing object attribute, aiming to achieve better object-level visual-language alignment and multimodal scene understanding. Specifically, we design an attribute fusion module and a contrastive knowledge distillation module. The attribute fusion module constructs a multimodal graph neural network to fuse attributes and visual features through message passing. The enhanced object-level visual features contribute to solving fine-grained problem like counting-question. The better object-level visual-language alignment aids in understanding multimodal scenes, thereby improving the model's robustness. Furthermore, to augment scene understanding and the out-of-distribution performance, the contrastive knowledge distillation module introduces a series of implicit knowledge. We distill knowledge into attributes through contrastive loss, which further strengthens the representation learning of attribute features and facilitates visual-linguistic alignment. Intensive experiments on six datasets, COCO-QA, VQAv2, VQA-CPv2, VQA-CPv1, VQAvs and TDIUC, show the superiority of the proposed method.

arXiv:2401.10139공식 코드 · Papers with Code

Model Compression Techniques in Biometrics Applications: A Survey

The development of deep learning algorithms has extensively empowered humanity's task automatization capacity. However, the huge improvement in the performance of these models is highly correlated with their increasing level of complexity, limiting their usefulness in human-oriented applications, which are usually deployed in resource-constrained devices. This led to the development of compression techniques that drastically reduce the computational and memory costs of deep learning models without significant performance degradation. This paper aims to systematize the current literature on this topic by presenting a comprehensive survey of model compression techniques in biometrics applications, namely quantization, knowledge distillation and pruning. We conduct a critical analysis of the comparative value of these techniques, focusing on their advantages and disadvantages and presenting suggestions for future work directions that can potentially improve the current methods. Additionally, we discuss and analyze the link between model bias and model compression, highlighting the need to direct compression research toward model fairness in future works.

arXiv:2401.08860

Cross-Level Multi-Instance Distillation for Self-Supervised Fine-Grained Visual Categorization

High-quality annotation of fine-grained visual categories demands great expert knowledge, which is taxing and time consuming. Alternatively, learning fine-grained visual representation from enormous unlabeled images (e.g., species, brands) by self-supervised learning becomes a feasible solution. However, recent researches find that existing self-supervised learning methods are less qualified to represent fine-grained categories. The bottleneck lies in that the pre-text representation is built from every patch-wise embedding, while fine-grained categories are only determined by several key patches of an image. In this paper, we propose a Cross-level Multi-instance Distillation (CMD) framework to tackle the challenge. Our key idea is to consider the importance of each image patch in determining the fine-grained pre-text representation by multiple instance learning. To comprehensively learn the relation between informative patches and fine-grained semantics, the multi-instance knowledge distillation is implemented on both the region/image crop pairs from the teacher and student net, and the region-image crops inside the teacher / student net, which we term as intra-level multi-instance distillation and inter-level multi-instance distillation. Extensive experiments on CUB-200-2011, Stanford Cars and FGVC Aircraft show that the proposed method outperforms the contemporary method by upto 10.14% and existing state-of-the-art self-supervised learning approaches by upto 19.78% on both top-1 accuracy and Rank-1 retrieval metric.

arXiv:2401.08732공식 코드 · Papers with Code

Bayes Conditional Distribution Estimation for Knowledge Distillation Based on Conditional Mutual Information

It is believed that in knowledge distillation (KD), the role of the teacher is to provide an estimate for the unknown Bayes conditional probability distribution (BCPD) to be used in the student training process. Conventionally, this estimate is obtained by training the teacher using maximum log-likelihood (MLL) method. To improve this estimate for KD, in this paper we introduce the concept of conditional mutual information (CMI) into the estimation of BCPD and propose a novel estimator called the maximum CMI (MCMI) method. Specifically, in MCMI estimation, both the log-likelihood and CMI of the teacher are simultaneously maximized when the teacher is trained. Through Eigen-CAM, it is further shown that maximizing the teacher's CMI value allows the teacher to capture more contextual information in an image cluster. Via conducting a thorough set of experiments, we show that by employing a teacher trained via MCMI estimation rather than one trained via MLL estimation in various state-of-the-art KD frameworks, the student's classification accuracy consistently increases, with the gain of up to 3.32\%. This suggests that the teacher's BCPD estimate provided by MCMI method is more accurate than that provided by MLL method. In addition, we show that such improvements in the student's accuracy are more drastic in zero-shot and few-shot settings. Notably, the student's accuracy increases with the gain of up to 5.72\% when 5\% of the training samples are available to the student (few-shot), and increases from 0\% to as high as 84\% for an omitted class (zero-shot). The code is available at \url{https://github.com/iclr2024mcmi/ICLRMCMI}.

arXiv:2401.07257

Lightweight Modality Adaptation to Sequential Recommendation via Correlation Supervision

In Sequential Recommenders (SR), encoding and utilizing modalities in an end-to-end manner is costly in terms of modality encoder sizes. Two-stage approaches can mitigate such concerns, but they suffer from poor performance due to modality forgetting, where the sequential objective overshadows modality representation. We propose a lightweight knowledge distillation solution that preserves both merits: retaining modality information and maintaining high efficiency. Specifically, we introduce a novel method that enhances the learning of embeddings in SR through the supervision of modality correlations. The supervision signals are distilled from the original modality representations, including both (1) holistic correlations, which quantify their overall associations, and (2) dissected correlation types, which refine their relationship facets (honing in on specific aspects like color or shape consistency). To further address the issue of modality forgetting, we propose an asynchronous learning step, allowing the original information to be retained longer for training the representation learning module. Our approach is compatible with various backbone architectures and outperforms the top baselines by 6.8% on average. We empirically demonstrate that preserving original feature associations from modality encoders significantly boosts task-specific recommendation adaptation. Additionally, we find that larger modality encoders (e.g., Large Language Models) contain richer feature sets which necessitate more fine-grained modeling to reach their full performance potential.

arXiv:2401.07013

Knowledge Distillation of Black-Box Large Language Models

Given the exceptional performance of proprietary large language models (LLMs) like GPT-4, recent research has increasingly focused on boosting the capabilities of smaller models through knowledge distillation (KD) from these powerful yet black-box teachers. While leveraging the high-quality outputs of these teachers is advantageous, the inaccessibility of their internal states often limits effective knowledge transfer. To overcome this limitation, we introduce Proxy-KD, a novel method that uses a proxy model to facilitate the efficient transfer of knowledge from black-box LLMs to smaller models. Our experiments show that Proxy-KD not only enhances the performance of KD from black-box teacher models but also surpasses traditional white-box KD techniques.~This approach presents a compelling new avenue for distilling knowledge from advanced LLMs.

arXiv:2401.06802

Hierarchical Knowledge Distillation on Text Graph for Data-limited Attribute Inference

The popularization of social media increases user engagements and generates a large amount of user-oriented data. Among them, text data (e.g., tweets, blogs) significantly attracts researchers and speculators to infer user attributes (e.g., age, gender, location) for fulfilling their intents. Generally, this line of work casts attribute inference as a text classification problem, and starts to leverage graph neural networks (GNNs) to utilize higher-level representations of source texts. However, these text graphs are constructed over words, suffering from high memory consumption and ineffectiveness on few labeled texts. To address this challenge, we design a text-graph-based few-shot learning model for attribute inferences on social media text data. Our model first constructs and refines a text graph using manifold learning and message passing, which offers a better trade-off between expressiveness and complexity. Afterwards, to further use cross-domain texts and unlabeled texts to improve few-shot performance, a hierarchical knowledge distillation is devised over text graph to optimize the problem, which derives better text representations, and advances model generalization ability. Experiments on social media datasets demonstrate the state-of-the-art performance of our model on attribute inferences with considerably fewer labeled texts.

arXiv:2401.07469

A Deep Hierarchical Feature Sparse Framework for Occluded Person Re-Identification

Most existing methods tackle the problem of occluded person re-identification (ReID) by utilizing auxiliary models, resulting in a complicated and inefficient ReID framework that is unacceptable for real-time applications. In this work, a speed-up person ReID framework named SUReID is proposed to mitigate occlusion interference while speeding up inference. The SUReID consists of three key components: hierarchical token sparsification (HTS) strategy, non-parametric feature alignment knowledge distillation (NPKD), and noise occlusion data augmentation (NODA). The HTS strategy works by pruning the redundant tokens in the vision transformer to achieve highly effective self-attention computation and eliminate interference from occlusions or background noise. However, the pruned tokens may contain human part features that contaminate the feature representation and degrade the performance. To solve this problem, the NPKD is employed to supervise the HTS strategy, retaining more discriminative tokens and discarding meaningless ones. Furthermore, the NODA is designed to introduce more noisy samples, which further trains the ability of the HTS to disentangle different tokens. Experimental results show that the SUReID achieves superior performance with surprisingly fast inference.

arXiv:2401.06957

EVOKE: Emotion Enabled Virtual Avatar Mapping Using Optimized Knowledge Distillation

As virtual environments continue to advance, the demand for immersive and emotionally engaging experiences has grown. Addressing this demand, we introduce Emotion enabled Virtual avatar mapping using Optimized KnowledgE distillation (EVOKE), a lightweight emotion recognition framework designed for the seamless integration of emotion recognition into 3D avatars within virtual environments. Our approach leverages knowledge distillation involving multi-label classification on the publicly available DEAP dataset, which covers valence, arousal, and dominance as primary emotional classes. Remarkably, our distilled model, a CNN with only two convolutional layers and 18 times fewer parameters than the teacher model, achieves competitive results, boasting an accuracy of 87% while demanding far less computational resources. This equilibrium between performance and deployability positions our framework as an ideal choice for virtual environment systems. Furthermore, the multi-label classification outcomes are utilized to map emotions onto custom-designed 3D avatars.

arXiv:2401.06826

Direct Distillation between Different Domains

Knowledge Distillation (KD) aims to learn a compact student network using knowledge from a large pre-trained teacher network, where both networks are trained on data from the same distribution. However, in practical applications, the student network may be required to perform in a new scenario (i.e., the target domain), which usually exhibits significant differences from the known scenario of the teacher network (i.e., the source domain). The traditional domain adaptation techniques can be integrated with KD in a two-stage process to bridge the domain gap, but the ultimate reliability of two-stage approaches tends to be limited due to the high computational consumption and the additional errors accumulated from both stages. To solve this problem, we propose a new one-stage method dubbed ``Direct Distillation between Different Domains" (4Ds). We first design a learnable adapter based on the Fourier transform to separate the domain-invariant knowledge from the domain-specific knowledge. Then, we build a fusion-activation mechanism to transfer the valuable domain-invariant knowledge to the student network, while simultaneously encouraging the adapter within the teacher network to learn the domain-specific knowledge of the target data. As a result, the teacher network can effectively transfer categorical knowledge that aligns with the target domain of the student network. Intensive experiments on various benchmark datasets demonstrate that our proposed 4Ds method successfully produces reliable student networks and outperforms state-of-the-art approaches.

arXiv:2401.08174공식 코드 · Papers with Code

OBSeg: Accurate and Fast Instance Segmentation Framework Using Segmentation Foundation Models with Oriented Bounding Box Prompts

Instance segmentation in remote sensing images is a long-standing challenge. Since horizontal bounding boxes introduce many interference objects, oriented bounding boxes (OBBs) are usually used for instance identification. However, based on ``segmentation within bounding box'' paradigm, current instance segmentation methods using OBBs are overly dependent on bounding box detection performance. To tackle this problem, this paper proposes OBSeg, an accurate and fast instance segmentation framework using OBBs. OBSeg is based on box prompt-based segmentation foundation models (BSMs), e.g., Segment Anything Model. Specifically, OBSeg first detects OBBs to distinguish instances and provide coarse localization information. Then, it predicts OBB prompt-related masks for fine segmentation. Since OBBs only serve as prompts, OBSeg alleviates the over-dependence on bounding box detection performance of current instance segmentation methods using OBBs. Thanks to OBB prompts, OBSeg outperforms other current BSM-based methods using HBBs. In addition, to enable BSMs to handle OBB prompts, we propose a novel OBB prompt encoder. To make OBSeg more lightweight and further improve the performance of lightweight distilled BSMs, a Gaussian smoothing-based knowledge distillation method is introduced. Experiments demonstrate that OBSeg outperforms current instance segmentation methods on multiple datasets in terms of instance segmentation accuracy and has competitive inference speed. The code is available at https://github.com/zhen6618/OBBInstanceSegmentation.

arXiv:2401.08332공식 코드 · Papers with Code

Generative Denoise Distillation: Simple Stochastic Noises Induce Efficient Knowledge Transfer for Dense Prediction

Knowledge distillation is the process of transferring knowledge from a more powerful large model (teacher) to a simpler counterpart (student). Numerous current approaches involve the student imitating the knowledge of the teacher directly. However, redundancy still exists in the learned representations through these prevalent methods, which tend to learn each spatial location's features indiscriminately. To derive a more compact representation (concept feature) from the teacher, inspired by human cognition, we suggest an innovative method, termed Generative Denoise Distillation (GDD), where stochastic noises are added to the concept feature of the student to embed them into the generated instance feature from a shallow network. Then, the generated instance feature is aligned with the knowledge of the instance from the teacher. We extensively experiment with object detection, instance segmentation, and semantic segmentation to demonstrate the versatility and effectiveness of our method. Notably, GDD achieves new state-of-the-art performance in the tasks mentioned above. We have achieved substantial improvements in semantic segmentation by enhancing PspNet and DeepLabV3, both of which are based on ResNet-18, resulting in mIoU scores of 74.67 and 77.69, respectively, surpassing their previous scores of 69.85 and 73.20 on the Cityscapes dataset of 20 categories. The source code is available at https://github.com/ZhgLiu/GDD.

arXiv:2401.06370공식 코드 · Papers with Code

Graph Relation Distillation for Efficient Biomedical Instance Segmentation

Instance-aware embeddings predicted by deep neural networks have revolutionized biomedical instance segmentation, but its resource requirements are substantial. Knowledge distillation offers a solution by transferring distilled knowledge from heavy teacher networks to lightweight yet high-performance student networks. However, existing knowledge distillation methods struggle to extract knowledge for distinguishing instances and overlook global relation information. To address these challenges, we propose a graph relation distillation approach for efficient biomedical instance segmentation, which considers three essential types of knowledge: instance-level features, instance relations, and pixel-level boundaries. We introduce two graph distillation schemes deployed at both the intra-image level and the inter-image level: instance graph distillation (IGD) and affinity graph distillation (AGD). IGD constructs a graph representing instance features and relations, transferring these two types of knowledge by enforcing instance graph consistency. AGD constructs an affinity graph representing pixel relations to capture structured knowledge of instance boundaries, transferring boundary-related knowledge by ensuring pixel affinity consistency. Experimental results on a number of biomedical datasets validate the effectiveness of our approach, enabling student models with less than $ 1\%$ parameters and less than $10\%$ inference time while achieving promising performance compared to teacher models.