The rapid integration of Internet of Things(IoT)technologies is reshaping the global energy landscape by deploying smart meters that enable high-resolution consumption monitoring,two-way communication,and advanced met...The rapid integration of Internet of Things(IoT)technologies is reshaping the global energy landscape by deploying smart meters that enable high-resolution consumption monitoring,two-way communication,and advanced metering infrastructure services.However,this digital transformation also exposes power system to evolving threats,ranging from cyber intrusions and electricity theft to device malfunctions,and the unpredictable nature of these anomalies,coupled with the scarcity of labeled fault data,makes realtime detection exceptionally challenging.To address these difficulties,a real-time decision support framework is presented for smart meter anomality detection that leverages rolling time windows and two self-supervised contrastive learning modules.The first module synthesizes diverse negative samples to overcome the lack of labeled anomalies,while the second captures intrinsic temporal patterns for enhanced contextual discrimination.The end-to-end framework continuously updates its model with rolling updated meter data to deliver timely identification of emerging abnormal behaviors in evolving grids.Extensive evaluations on eight publicly available smart meter datasets over seven diverse abnormal patterns testing demonstrate the effectiveness of the proposed full framework,achieving average recall and F1 score of more than 0.85.展开更多
To address the issue of scarce labeled samples and operational condition variations that degrade the accuracy of fault diagnosis models in variable-condition gearbox fault diagnosis,this paper proposes a semi-supervis...To address the issue of scarce labeled samples and operational condition variations that degrade the accuracy of fault diagnosis models in variable-condition gearbox fault diagnosis,this paper proposes a semi-supervised masked contrastive learning and domain adaptation(SSMCL-DA)method for gearbox fault diagnosis under variable conditions.Initially,during the unsupervised pre-training phase,a dual signal augmentation strategy is devised,which simultaneously applies random masking in the time domain and random scaling in the frequency domain to unlabeled samples,thereby constructing more challenging positive sample pairs to guide the encoder in learning intrinsic features robust to condition variations.Subsequently,a ConvNeXt-Transformer hybrid architecture is employed,integrating the superior local detail modeling capacity of ConvNeXt with the robust global perception capability of Transformer to enhance feature extraction in complex scenarios.Thereafter,a contrastive learning model is constructed with the optimization objective of maximizing feature similarity across different masked instances of the same sample,enabling the extraction of consistent features from multiple masked perspectives and reducing reliance on labeled data.In the final supervised fine-tuning phase,a multi-scale attention mechanism is incorporated for feature rectification,and a domain adaptation module combining Local Maximum Mean Discrepancy(LMMD)with adversarial learning is proposed.This module embodies a dual mechanism:LMMD facilitates fine-grained class-conditional alignment,compelling features of identical fault classes to converge across varying conditions,while the domain discriminator utilizes adversarial training to guide the feature extractor toward learning domain-invariant features.Working in concert,they markedly diminish feature distribution discrepancies induced by changes in load,rotational speed,and other factors,thereby boosting the model’s adaptability to cross-condition scenarios.Experimental evaluations on the WT planetary gearbox dataset and the Case Western Reserve University(CWRU)bearing dataset demonstrate that the SSMCL-DA model effectively identifies multiple fault classes in gearboxes,with diagnostic performance substantially surpassing that of conventional methods.Under cross-condition scenarios,the model attains fault diagnosis accuracies of 99.21%for the WT planetary gearbox and 99.86%for the bearings,respectively.Furthermore,the model exhibits stable generalization capability in cross-device settings.展开更多
Two-dimensional endoscopic images are susceptible to interferences such as specular reflections and monotonous texture illumination,hindering accurate three-dimensional lesion reconstruction by surgical robots.This st...Two-dimensional endoscopic images are susceptible to interferences such as specular reflections and monotonous texture illumination,hindering accurate three-dimensional lesion reconstruction by surgical robots.This study proposes a novel end-to-end disparity estimation model to address these challenges.Our approach combines a Pseudo-Siamese neural network architecture with pyramid dilated convolutions,integrating multi-scale image information to enhance robustness against lighting interferences.This study introduces a Pseudo-Siamese structure-based disparity regression model that simplifies left-right image comparison,improving accuracy and efficiency.The model was evaluated using a dataset of stereo endoscopic videos captured by the Da Vinci surgical robot,comprising simulated silicone heart sequences and real heart video data.Experimental results demonstrate significant improvement in the network’s resistance to lighting interference without substantially increasing parameters.Moreover,the model exhibited faster convergence during training,contributing to overall performance enhancement.This study advances endoscopic image processing accuracy and has potential implications for surgical robot applications in complex environments.展开更多
Intelligent Transportation Systems(ITS)leverage Integrated Sensing and Communications(ISAC)to enhance data exchange between vehicles and infrastructure in the Internet of Vehicles(IoV).This integration inevitably incr...Intelligent Transportation Systems(ITS)leverage Integrated Sensing and Communications(ISAC)to enhance data exchange between vehicles and infrastructure in the Internet of Vehicles(IoV).This integration inevitably increases computing demands,risking real-time system stability.Vehicle Edge Computing(VEC)addresses this by offloading tasks to Road Side Units(RSUs),ensuring timely services.Our previous work,the FLSimCo algorithm,which uses local resources for federated Self-Supervised Learning(SSL),has a limitation:vehicles often can’t complete all iteration tasks.Our improved algorithm offloads partial tasks to RSUs and optimizes energy consumption by adjusting transmission power,CPU frequency,and task assignment ratios,balancing local and RSU-based training.Meanwhile,setting an offloading threshold further prevents inefficiencies.Simulation results show that the enhanced algorithm reduces energy consumption and improves offloading efficiency and accuracy of federated SSL.展开更多
Learning discriminative representations with deep neural networks often relies on massive labeled data, which is expensive and difficult to obtain in many real scenarios. As an alternative, self-supervised learning th...Learning discriminative representations with deep neural networks often relies on massive labeled data, which is expensive and difficult to obtain in many real scenarios. As an alternative, self-supervised learning that leverages input itself as supervision is strongly preferred for its soaring performance on visual representation learning. This paper introduces a contrastive self-supervised framework for learning generalizable representations on the synthetic data that can be obtained easily with complete controllability.Specifically, we propose to optimize a contrastive learning task and a physical property prediction task simultaneously. Given the synthetic scene, the first task aims to maximize agreement between a pair of synthetic images generated by our proposed view sampling module, while the second task aims to predict three physical property maps, i.e., depth, instance contour maps, and surface normal maps. In addition, a feature-level domain adaptation technique with adversarial training is applied to reduce the domain difference between the realistic and the synthetic data. Experiments demonstrate that our proposed method achieves state-of-the-art performance on several visual recognition datasets.展开更多
Automated grading of dandruff severity is a clinically significant but challenging task due to the inherent ordinal nature of severity levels and the high prevalence of label noise from subjective expert annotations.S...Automated grading of dandruff severity is a clinically significant but challenging task due to the inherent ordinal nature of severity levels and the high prevalence of label noise from subjective expert annotations.Standard classification methods fail to address these dual challenges,limiting their real-world performance.In this paper,a novel,three-phase training framework is proposed that learns a robust ordinal classifier directly from noisy labels.The approach synergistically combines a rank-based ordinal regression backbone with a cooperative,semi-supervised learning strategy to dynamically partition the data into clean and noisy subsets.A hybrid training objective is then employed,applying a supervised ordinal loss to the clean set.The noisy set is simultaneously trained using a dualobjective that combines a semi-supervised ordinal loss with a parallel,label-agnostic contrastive loss.This design allows themodel to learn fromthe entire noisy subset while using contrastive learning to mitigate the risk of error propagation frompotentially corrupt supervision.Extensive experiments on a new,large-scale,multi-site clinical dataset validate our approach.Themethod achieves state-of-the-art performance with 80.71%accuracy and a 76.86%F1-score,significantly outperforming existing approaches,including a 2.26%improvement over the strongest baseline method.This work provides not only a robust solution for a practical medical imaging problem but also a generalizable framework for other tasks plagued by noisy ordinal labels.展开更多
Wearable wristband systems leverage deep learning to revolutionize hand gesture recognition in daily activities.Unlike existing approaches that often focus on static gestures and require extensive labeled data,the pro...Wearable wristband systems leverage deep learning to revolutionize hand gesture recognition in daily activities.Unlike existing approaches that often focus on static gestures and require extensive labeled data,the proposed wearable wristband with selfsupervised contrastive learning excels at dynamic motion tracking and adapts rapidly across multiple scenarios.It features a four-channel sensing array composed of an ionic hydrogel with hierarchical microcone structures and ultrathin flexible electrodes,resulting in high-sensitivity capacitance output.Through wireless transmission from a Wi-Fi module,the proposed algorithm learns latent features from the unlabeled signals of random wrist movements.Remarkably,only few-shot labeled data are sufficient for fine-tuning the model,enabling rapid adaptation to various tasks.The system achieves a high accuracy of 94.9%in different scenarios,including the prediction of eight-direction commands,and air-writing of all numbers and letters.The proposed method facilitates smooth transitions between multiple tasks without the need for modifying the structure or undergoing extensive task-specific training.Its utility has been further extended to enhance human–machine interaction over digital platforms,such as game controls,calculators,and three-language login systems,offering users a natural and intuitive way of communication.展开更多
The collection and annotation of lar ge-scale bird datasets are resource-intensive and time-consuming processes that significantly limit the scalability and accuracy of biodiversity monitoring systems.While self-super...The collection and annotation of lar ge-scale bird datasets are resource-intensive and time-consuming processes that significantly limit the scalability and accuracy of biodiversity monitoring systems.While self-supervised learning(SSL)has emerged as a promising approach for leveraging unannotated data,current SSL methods face two critical challenges in bird species recognition:(1)long-tailed data distributions that result in poor performance on underrepresented species;and(2)domain shift issues caused by data augmentation strategies designed to mitigate class imbalance.Here we present SDNet,a novel SSL-based bird recognition framework that integrates diffusion models with large language models(LLMs)to overcome these limitations.SDNet employs LLMs to generate semantically rich textual descriptions for tail-class species by prompting the models with species taxonomy,morphological attributes,and habitat information,producing detailed natural language priors that capture fine-grained visual characteristics(e.g.,plumage patterns,body proportions,and distinctive markings).These textual descriptions are subsequently used by a conditional diffusion model to synthesize new bird image samples through cross-attention mechanisms that fuse textual embeddings with intermediate visual feature representations during the denoising process,ensuring generated images preserve species-specific morphological details while maintaining photorealistic quality.Additionally,we incorporate a Swin Transformer as the feature extraction backbone whose hierarchical window-based attention mechanism and shifted windowing scheme enable multi-scale local feature extraction that proves particularly effective at capturing finegrained discriminative patterns(such as beak shape and feather texture)while mitigating domain shift between synthetic and original images through consistent feature representations across both data sources.SDNet is validated on both a self-constructed dataset(Bird_BXS)an d a publicly available benchmark(Birds_25),demonstrating substantial improvements over conventional SSL approaches.Our results indicate that the synergistic integration of LLMs,diffusion models,and the Swin Transformer architecture contributes significantly to recognition accuracy,particularly for rare and morphologically similar species.These findings highlight the potential of SDNet for addressing fundamental limitations of existing SSL methods in avian recognition tasks and establishing a new paradigm for efficient self-supervised learning in large-scale ornithological vision applications.展开更多
The accurate segmentation of deep gray matter nuclei is critical for neuropathological research,disease diagnosis and treatment.Existing methods employ the supervised learning training approach,which requires large la...The accurate segmentation of deep gray matter nuclei is critical for neuropathological research,disease diagnosis and treatment.Existing methods employ the supervised learning training approach,which requires large labeled datasets.It is challenging and time-consuming to obtain such datasets for medical image analysis.In addition,these methods based on convolutional neural networks(CNNs)only achieve suboptimal performance due to the locality of convolutional operations.Vision Transformers(ViTs)efficiently model long-range dependencies and thus have the potentiality to outperform these methods in segmentation tasks.To address these issues,we propose a novel hybrid network based on self-supervised pre-training for deep gray matter nuclei segmentation.Specifically,we present a CNN-Transformer hybrid network(CTNet),whose encoder consists of 3D CNN and ViT to learn local spatial-detailed features and global semantic information.A self-supervised learning(SSL)approach that integrates rotation prediction and masked feature reconstruction is proposed to pre-train the CTNet,enabling the model to learn valuable visual representations from unlabeled data.We evaluate the effectiveness of our method on 3T and 7T human brain MRI datasets.The results demonstrate that our CTNet achieves better performance than other comparison models and our pre-training strategy outperforms other advanced self-supervised methods.When the training set has only one sample,our pre-trained CTNet enhances segmentation performance,showing an 8.4%improvement in Dice similarity coefficient(DSC)compared to the randomly initialized CTNet.展开更多
P-glycoprotein(P-gp)is a transmembrane protein widely involved in the absorption,distribution,metabolism,excretion,and toxicity(ADMET)of drugs within the human body.Accurate prediction of Pgp inhibitors and substrates...P-glycoprotein(P-gp)is a transmembrane protein widely involved in the absorption,distribution,metabolism,excretion,and toxicity(ADMET)of drugs within the human body.Accurate prediction of Pgp inhibitors and substrates is crucial for drug discovery and toxicological assessment.However,existing models rely on limited molecular information,leading to suboptimal model performance for predicting P-gp inhibitors and substrates.To overcome this challenge,we compiled an extensive dataset from public databases and literature,consisting of 5,943 P-gp inhibitors and 4,018 substrates,notable for their high quantity,quality,and structural uniqueness.In addition,we curated two external test sets to validate the model's generalization capability.Subsequently,we developed a multimodal graph contrastive learning(GCL)model for the prediction of P-gp inhibitors and substrates(MC-PGP).This framework integrates three types of features from Simplified Molecular Input Line Entry System(SMILES)sequences,molecular fingerprints,and molecular graphs using an attention-based fusion strategy to generate a unified molecular representation.Furthermore,we employed a GCL approach to enhance structural representations by aligning local and global structures.Extensive experimental results highlight the superior performance of MC-PGP,which achieves improvements in the area under the curve of receiver operating characteristic(AUC-ROC)of 9.82%and 10.62%on the external P-gp inhibitor and external P-gp substrate datasets,respectively,compared with 12 state-of-the-art methods.Furthermore,the interpretability analysis of all three molecular feature types offers comprehensive and complementary insights,demonstrating that MC-PGP effectively identifies key functional groups involved in P-gp interactions.These chemically intuitive insights provide valuable guidance for the design and optimization of drug candidates.展开更多
Emotion recognition plays a crucial role in various fields and is a key task in natural language processing (NLP). The objective is to identify and interpret emotional expressions in text. However, traditional emotion...Emotion recognition plays a crucial role in various fields and is a key task in natural language processing (NLP). The objective is to identify and interpret emotional expressions in text. However, traditional emotion recognition approaches often struggle in few-shot cross-domain scenarios due to their limited capacity to generalize semantic features across different domains. Additionally, these methods face challenges in accurately capturing complex emotional states, particularly those that are subtle or implicit. To overcome these limitations, we introduce a novel approach called Dual-Task Contrastive Meta-Learning (DTCML). This method combines meta-learning and contrastive learning to improve emotion recognition. Meta-learning enhances the model’s ability to generalize to new emotional tasks, while instance contrastive learning further refines the model by distinguishing unique features within each category, enabling it to better differentiate complex emotional expressions. Prototype contrastive learning, in turn, helps the model address the semantic complexity of emotions across different domains, enabling the model to learn fine-grained emotions expression. By leveraging dual tasks, DTCML learns from two domains simultaneously, the model is encouraged to learn more diverse and generalizable emotions features, thereby improving its cross-domain adaptability and robustness, and enhancing its generalization ability. We evaluated the performance of DTCML across four cross-domain settings, and the results show that our method outperforms the best baseline by 5.88%, 12.04%, 8.49%, and 8.40% in terms of accuracy.展开更多
Federated learning(FL)is a distributed machine learning paradigm for edge cloud computing.FL can facilitate data-driven decision-making in tactical scenarios,effectively addressing both data volume and infrastructure ...Federated learning(FL)is a distributed machine learning paradigm for edge cloud computing.FL can facilitate data-driven decision-making in tactical scenarios,effectively addressing both data volume and infrastructure challenges in edge environments.However,the diversity of clients in edge cloud computing presents significant challenges for FL.Personalized federated learning(pFL)received considerable attention in recent years.One example of pFL involves exploiting the global and local information in the local model.Current pFL algorithms experience limitations such as slow convergence speed,catastrophic forgetting,and poor performance in complex tasks,which still have significant shortcomings compared to the centralized learning.To achieve high pFL performance,we propose FedCLCC:Federated Contrastive Learning and Conditional Computing.The core of FedCLCC is the use of contrastive learning and conditional computing.Contrastive learning determines the feature representation similarity to adjust the local model.Conditional computing separates the global and local information and feeds it to their corresponding heads for global and local handling.Our comprehensive experiments demonstrate that FedCLCC outperforms other state-of-the-art FL algorithms.展开更多
Graph similarity learning aims to calculate the similarity between pairs of graphs.Existing unsupervised graph similarity learning methods based on contrastive learning encounter challenges related to random graph aug...Graph similarity learning aims to calculate the similarity between pairs of graphs.Existing unsupervised graph similarity learning methods based on contrastive learning encounter challenges related to random graph augmentation strategies,which can harm the semantic and structural information of graphs and overlook the rich structural information present in subgraphs.To address these issues,we propose a graph similarity learning model based on learnable augmentation and multi-level contrastive learning.First,to tackle the problem of random augmentation disrupting the semantics and structure of the graph,we design a learnable augmentation method to selectively choose nodes and edges within the graph.To enhance contrastive levels,we employ a biased random walk method to generate corresponding subgraphs,enriching the contrastive hierarchy.Second,to solve the issue of previous work not considering multi-level contrastive learning,we utilize graph convolutional networks to learn node representations of augmented views and the original graph and calculate the interaction information between the attribute-augmented and structure-augmented views and the original graph.The goal is to maximize node consistency between different views and learn node matching between different graphs,resulting in node-level representations for each graph.Subgraph representations are then obtained through pooling operations,and we conduct contrastive learning utilizing both node and subgraph representations.Finally,the graph similarity score is computed according to different downstream tasks.We conducted three sets of experiments across eight datasets,and the results demonstrate that the proposed model effectively mitigates the issues of random augmentation damaging the original graph’s semantics and structure,as well as the insufficiency of contrastive levels.Additionally,the model achieves the best overall performance.展开更多
In the big data era,the surge in network traffic volume poses challenges for network management and cybersecurity.Network Traffic Classification(NTC)employs deep learning to categorize traffic data,aiding security and...In the big data era,the surge in network traffic volume poses challenges for network management and cybersecurity.Network Traffic Classification(NTC)employs deep learning to categorize traffic data,aiding security and analysis systems as Intrusion Detection Systems(IDS)and Intrusion Prevention Systems(IPS).However,current NTC methods,based on isolated network simulations,usually fail to adapt to new protocols and applications and ignore the effects of network conditions and user behavior on traffic patterns.To improve network traffic management insights,federated learning frameworks have been proposed to aggregate diverse traffic data for collaborative model training.This approach faces challenges like data integrity,label noise,packet loss,and skewed data distributions.While label noise can be mitigated through the use of sophisticated traffic labeling tools,other issues such as packet loss and skewed data distributions encountered in Network Packet Brokers(NPB)can severely impede the efficacy of federated learning algorithms.In this paper,we introduced the Robust Traffic Classifier with Federated Contrastive Learning(FC-RTC),combining federated and contrastive learning methods.Using the Supcon-Loss function from contrastive learning,FC-RTC distinguishes between similar and dissimilar samples.Training by sample pairs,FC-RTC effectively updates when receiving corrupted traffic data with packet loss or disorder.In cases of sample imbalance,contrastive loss functions for similar samples reduce model bias towards higher proportion data.By addressing uneven data distribution and packet loss,our system enhances its capability to adapt and perform accurately in real-world network traffic analysis,meeting the specific demands of this complex field.展开更多
Fisheye cameras offer a significantly larger field of view compared to conventional cameras,making them valuable tools in the field of computer vision.However,their unique optical characteristics often lead to image d...Fisheye cameras offer a significantly larger field of view compared to conventional cameras,making them valuable tools in the field of computer vision.However,their unique optical characteristics often lead to image distortions,which pose challenges for object detection tasks.To address this issue,we propose Yolo-CaSKA(Yolo with Contrastive Learning and Selective Kernel Attention),a novel training method that enhances object detection on fisheye camera images.The standard image and the corresponding distorted fisheye image pairs are used as positive samples,and the rest of the image pairs are used as negative samples,which are guided by contrastive learning to help the distorted images find the feature vectors of the corresponding normal images,to improve the detection accuracy.Additionally,we incorporate the Selective Kernel(SK)attention module to focus on regions prone to false detections,such as image edges and blind spots.Finally,the mAP_(50) on the augmented KITTI dataset is improved by 5.5% over the original Yolov8,while the mAP_(50) on the WoodScape dataset is improved by 2.6% compared to OmniDet.The results demonstrate the performance of our proposed model for object detection on fisheye images.展开更多
Although conventional object detection methods achieve high accuracy through extensively annotated datasets,acquiring such large-scale labeled data remains challenging and cost-prohibitive in numerous real-world appli...Although conventional object detection methods achieve high accuracy through extensively annotated datasets,acquiring such large-scale labeled data remains challenging and cost-prohibitive in numerous real-world applications.Few-shot object detection presents a new research idea that aims to localize and classify objects in images using only limited annotated examples.However,the inherent challenge in few-shot object detection lies in the insufficient sample diversity to fully characterize the sample feature distribution,which consequently impacts model performance.Inspired by contrastive learning principles,we propose an Implicit Feature Contrastive Learning(IFCL)module to address this limitation and augment feature diversity for more robust representational learning.This module generates augmented support sample features in a mixed feature space and implicitly contrasts them with query Region of Interest(RoI)features.This approach facilitates more comprehensive learning of both intra-class feature similarity and inter-class feature diversity,thereby enhancing the model’s object classification and localization capabilities.Extensive experiments on PASCAL VOC show that our method achieves a respective improvement of 3.2%,1.8%,and 2.3%on 10-shot of three Novel Sets compared to the baseline model FPD.展开更多
Sarcasm detection is a complex and challenging task,particularly in the context of Chinese social media,where it exhibits strong contextual dependencies and cultural specificity.To address the limitations of existing ...Sarcasm detection is a complex and challenging task,particularly in the context of Chinese social media,where it exhibits strong contextual dependencies and cultural specificity.To address the limitations of existing methods in capturing the implicit semantics and contextual associations in sarcastic expressions,this paper proposes an event-aware model for Chinese sarcasm detection,leveraging a multi-head attention(MHA)mechanism and contrastive learning(CL)strategies.The proposed model employs a dual-path Bidirectional Encoder Representations from Transformers(BERT)encoder to process comment text and event context separately and integrates an MHA mechanism to facilitate deep interactions between the two,thereby capturing multidimensional semantic associations.Additionally,a CL strategy is introduced to enhance feature representation capabilities,further improving the model’s performance in handling class imbalance and complex contextual scenarios.The model achieves state-of-the-art performance on the Chinese sarcasm dataset,with significant improvements in accuracy(79.55%),F1-score(84.22%),and an area under the curve(AUC,84.35%).展开更多
In the field of optoelectronics,certain types of data may be difficult to accurately annotate,such as high-resolution optoelectronic imaging or imaging in certain special spectral ranges.Weakly supervised learning can...In the field of optoelectronics,certain types of data may be difficult to accurately annotate,such as high-resolution optoelectronic imaging or imaging in certain special spectral ranges.Weakly supervised learning can provide a more reliable approach in these situations.Current popular approaches mainly adopt the classification-based class activation maps(CAM)as initial pseudo labels to solve the task.展开更多
In federated learning(FL),the distribution of data across different clients leads to the degradation of global model performance in training.Personalized Federated Learning(pFL)can address this problem through global ...In federated learning(FL),the distribution of data across different clients leads to the degradation of global model performance in training.Personalized Federated Learning(pFL)can address this problem through global model personalization.Researches over the past few years have calibrated differences in weights across the entire model or optimized only individual layers of the model without considering that different layers of the whole neural network have different utilities,resulting in lagged model convergence and inadequate personalization in non-IID data.In this paper,we propose model layered optimization for feature extractor and classifier(pFedEC),a novel pFL training framework personalized for different layers of the model.Our study divides the model layers into the feature extractor and classifier.We initialize the model's classifiers during model training,while making the local model's feature extractors learn the representation of the global model's feature extractors to correct each client's local training,integrating the utilities of the different layers in the entire model.Our extensive experiments show that pFedEC achieves 92.95%accuracy on CIFAR-10,outperforming existing pFL methods by approximately 1.8%.On CIFAR-100 and Tiny-ImageNet,pFedEC improves the accuracy by at least 4.2%,reaching 73.02%and 28.39%,respectively.展开更多
Few-shot point cloud 3D object detection(FS3D)aims to identify and locate objects of novel classes within point clouds using knowledge acquired from annotated base classes and a minimal number of samples from the nove...Few-shot point cloud 3D object detection(FS3D)aims to identify and locate objects of novel classes within point clouds using knowledge acquired from annotated base classes and a minimal number of samples from the novel classes.Due to imbalanced training data,existing FS3D methods based on fully supervised learning can lead to overfitting toward base classes,which impairs the network’s ability to generalize knowledge learned from base classes to novel classes and also prevents the network from extracting distinctive foreground and background representations for novel class objects.To address these issues,this thesis proposes a category-agnostic contrastive learning approach,enhancing the generalization and identification abilities for almost unseen categories through the construction of pseudo-labels and positive-negative sample pairs unrelated to specific classes.Firstly,this thesis designs a proposal-wise context contrastive module(CCM).By reducing the distance between foreground point features and increasing the distance between foreground and background point features within a region proposal,CCM aids the network in extracting more discriminative foreground and background feature representations without reliance on categorical annotations.Secondly,this thesis utilizes a geometric contrastive module(GCM),which enhances the network’s geometric perception capability by employing contrastive learning on the foreground point features associated with various basic geometric components,such as edges,corners,and surfaces,thereby enabling these geometric components to exhibit more distinguishable representations.This thesis also combines category-aware contrastive learning with former modules to maintain categorical distinctiveness.Extensive experimental results on FS-SUNRGBD and FS-ScanNet datasets demonstrate the effectiveness of this method with average precision exceeding the baseline by up to 8%.展开更多
文摘The rapid integration of Internet of Things(IoT)technologies is reshaping the global energy landscape by deploying smart meters that enable high-resolution consumption monitoring,two-way communication,and advanced metering infrastructure services.However,this digital transformation also exposes power system to evolving threats,ranging from cyber intrusions and electricity theft to device malfunctions,and the unpredictable nature of these anomalies,coupled with the scarcity of labeled fault data,makes realtime detection exceptionally challenging.To address these difficulties,a real-time decision support framework is presented for smart meter anomality detection that leverages rolling time windows and two self-supervised contrastive learning modules.The first module synthesizes diverse negative samples to overcome the lack of labeled anomalies,while the second captures intrinsic temporal patterns for enhanced contextual discrimination.The end-to-end framework continuously updates its model with rolling updated meter data to deliver timely identification of emerging abnormal behaviors in evolving grids.Extensive evaluations on eight publicly available smart meter datasets over seven diverse abnormal patterns testing demonstrate the effectiveness of the proposed full framework,achieving average recall and F1 score of more than 0.85.
基金supported by the National Natural Science Foundation of China Funded Project(Project Name:Research on Robust Adaptive Allocation Mechanism of Human Machine Co-Driving System Based on NMS Features,Project Approval Number:52172381).
文摘To address the issue of scarce labeled samples and operational condition variations that degrade the accuracy of fault diagnosis models in variable-condition gearbox fault diagnosis,this paper proposes a semi-supervised masked contrastive learning and domain adaptation(SSMCL-DA)method for gearbox fault diagnosis under variable conditions.Initially,during the unsupervised pre-training phase,a dual signal augmentation strategy is devised,which simultaneously applies random masking in the time domain and random scaling in the frequency domain to unlabeled samples,thereby constructing more challenging positive sample pairs to guide the encoder in learning intrinsic features robust to condition variations.Subsequently,a ConvNeXt-Transformer hybrid architecture is employed,integrating the superior local detail modeling capacity of ConvNeXt with the robust global perception capability of Transformer to enhance feature extraction in complex scenarios.Thereafter,a contrastive learning model is constructed with the optimization objective of maximizing feature similarity across different masked instances of the same sample,enabling the extraction of consistent features from multiple masked perspectives and reducing reliance on labeled data.In the final supervised fine-tuning phase,a multi-scale attention mechanism is incorporated for feature rectification,and a domain adaptation module combining Local Maximum Mean Discrepancy(LMMD)with adversarial learning is proposed.This module embodies a dual mechanism:LMMD facilitates fine-grained class-conditional alignment,compelling features of identical fault classes to converge across varying conditions,while the domain discriminator utilizes adversarial training to guide the feature extractor toward learning domain-invariant features.Working in concert,they markedly diminish feature distribution discrepancies induced by changes in load,rotational speed,and other factors,thereby boosting the model’s adaptability to cross-condition scenarios.Experimental evaluations on the WT planetary gearbox dataset and the Case Western Reserve University(CWRU)bearing dataset demonstrate that the SSMCL-DA model effectively identifies multiple fault classes in gearboxes,with diagnostic performance substantially surpassing that of conventional methods.Under cross-condition scenarios,the model attains fault diagnosis accuracies of 99.21%for the WT planetary gearbox and 99.86%for the bearings,respectively.Furthermore,the model exhibits stable generalization capability in cross-device settings.
基金Supported by Sichuan Science and Technology Program(2023YFSY0026,2023YFH0004)Supported by the Institute of Information&Communications Technology Planning&Evaluation(IITP)grant funded by the Korean government(MSIT)(No.RS-2022-00155885,Artificial Intelligence Convergence Innovation Human Resources Development(Hanyang University ERICA)).
文摘Two-dimensional endoscopic images are susceptible to interferences such as specular reflections and monotonous texture illumination,hindering accurate three-dimensional lesion reconstruction by surgical robots.This study proposes a novel end-to-end disparity estimation model to address these challenges.Our approach combines a Pseudo-Siamese neural network architecture with pyramid dilated convolutions,integrating multi-scale image information to enhance robustness against lighting interferences.This study introduces a Pseudo-Siamese structure-based disparity regression model that simplifies left-right image comparison,improving accuracy and efficiency.The model was evaluated using a dataset of stereo endoscopic videos captured by the Da Vinci surgical robot,comprising simulated silicone heart sequences and real heart video data.Experimental results demonstrate significant improvement in the network’s resistance to lighting interference without substantially increasing parameters.Moreover,the model exhibited faster convergence during training,contributing to overall performance enhancement.This study advances endoscopic image processing accuracy and has potential implications for surgical robot applications in complex environments.
文摘Intelligent Transportation Systems(ITS)leverage Integrated Sensing and Communications(ISAC)to enhance data exchange between vehicles and infrastructure in the Internet of Vehicles(IoV).This integration inevitably increases computing demands,risking real-time system stability.Vehicle Edge Computing(VEC)addresses this by offloading tasks to Road Side Units(RSUs),ensuring timely services.Our previous work,the FLSimCo algorithm,which uses local resources for federated Self-Supervised Learning(SSL),has a limitation:vehicles often can’t complete all iteration tasks.Our improved algorithm offloads partial tasks to RSUs and optimizes energy consumption by adjusting transmission power,CPU frequency,and task assignment ratios,balancing local and RSU-based training.Meanwhile,setting an offloading threshold further prevents inefficiencies.Simulation results show that the enhanced algorithm reduces energy consumption and improves offloading efficiency and accuracy of federated SSL.
基金by National Natural Science Foundation of China(Nos.61822204 and 61521002).
文摘Learning discriminative representations with deep neural networks often relies on massive labeled data, which is expensive and difficult to obtain in many real scenarios. As an alternative, self-supervised learning that leverages input itself as supervision is strongly preferred for its soaring performance on visual representation learning. This paper introduces a contrastive self-supervised framework for learning generalizable representations on the synthetic data that can be obtained easily with complete controllability.Specifically, we propose to optimize a contrastive learning task and a physical property prediction task simultaneously. Given the synthetic scene, the first task aims to maximize agreement between a pair of synthetic images generated by our proposed view sampling module, while the second task aims to predict three physical property maps, i.e., depth, instance contour maps, and surface normal maps. In addition, a feature-level domain adaptation technique with adversarial training is applied to reduce the domain difference between the realistic and the synthetic data. Experiments demonstrate that our proposed method achieves state-of-the-art performance on several visual recognition datasets.
文摘Automated grading of dandruff severity is a clinically significant but challenging task due to the inherent ordinal nature of severity levels and the high prevalence of label noise from subjective expert annotations.Standard classification methods fail to address these dual challenges,limiting their real-world performance.In this paper,a novel,three-phase training framework is proposed that learns a robust ordinal classifier directly from noisy labels.The approach synergistically combines a rank-based ordinal regression backbone with a cooperative,semi-supervised learning strategy to dynamically partition the data into clean and noisy subsets.A hybrid training objective is then employed,applying a supervised ordinal loss to the clean set.The noisy set is simultaneously trained using a dualobjective that combines a semi-supervised ordinal loss with a parallel,label-agnostic contrastive loss.This design allows themodel to learn fromthe entire noisy subset while using contrastive learning to mitigate the risk of error propagation frompotentially corrupt supervision.Extensive experiments on a new,large-scale,multi-site clinical dataset validate our approach.Themethod achieves state-of-the-art performance with 80.71%accuracy and a 76.86%F1-score,significantly outperforming existing approaches,including a 2.26%improvement over the strongest baseline method.This work provides not only a robust solution for a practical medical imaging problem but also a generalizable framework for other tasks plagued by noisy ordinal labels.
基金supported by the Research Grant Fund from Kwangwoon University in 2023,the National Natural Science Foundation of China under Grant(62311540155)the Taishan Scholars Project Special Funds(tsqn202312035)the open research foundation of State Key Laboratory of Integrated Chips and Systems.
文摘Wearable wristband systems leverage deep learning to revolutionize hand gesture recognition in daily activities.Unlike existing approaches that often focus on static gestures and require extensive labeled data,the proposed wearable wristband with selfsupervised contrastive learning excels at dynamic motion tracking and adapts rapidly across multiple scenarios.It features a four-channel sensing array composed of an ionic hydrogel with hierarchical microcone structures and ultrathin flexible electrodes,resulting in high-sensitivity capacitance output.Through wireless transmission from a Wi-Fi module,the proposed algorithm learns latent features from the unlabeled signals of random wrist movements.Remarkably,only few-shot labeled data are sufficient for fine-tuning the model,enabling rapid adaptation to various tasks.The system achieves a high accuracy of 94.9%in different scenarios,including the prediction of eight-direction commands,and air-writing of all numbers and letters.The proposed method facilitates smooth transitions between multiple tasks without the need for modifying the structure or undergoing extensive task-specific training.Its utility has been further extended to enhance human–machine interaction over digital platforms,such as game controls,calculators,and three-language login systems,offering users a natural and intuitive way of communication.
基金supported by the National Natural Science Foundation of China(32471964)。
文摘The collection and annotation of lar ge-scale bird datasets are resource-intensive and time-consuming processes that significantly limit the scalability and accuracy of biodiversity monitoring systems.While self-supervised learning(SSL)has emerged as a promising approach for leveraging unannotated data,current SSL methods face two critical challenges in bird species recognition:(1)long-tailed data distributions that result in poor performance on underrepresented species;and(2)domain shift issues caused by data augmentation strategies designed to mitigate class imbalance.Here we present SDNet,a novel SSL-based bird recognition framework that integrates diffusion models with large language models(LLMs)to overcome these limitations.SDNet employs LLMs to generate semantically rich textual descriptions for tail-class species by prompting the models with species taxonomy,morphological attributes,and habitat information,producing detailed natural language priors that capture fine-grained visual characteristics(e.g.,plumage patterns,body proportions,and distinctive markings).These textual descriptions are subsequently used by a conditional diffusion model to synthesize new bird image samples through cross-attention mechanisms that fuse textual embeddings with intermediate visual feature representations during the denoising process,ensuring generated images preserve species-specific morphological details while maintaining photorealistic quality.Additionally,we incorporate a Swin Transformer as the feature extraction backbone whose hierarchical window-based attention mechanism and shifted windowing scheme enable multi-scale local feature extraction that proves particularly effective at capturing finegrained discriminative patterns(such as beak shape and feather texture)while mitigating domain shift between synthetic and original images through consistent feature representations across both data sources.SDNet is validated on both a self-constructed dataset(Bird_BXS)an d a publicly available benchmark(Birds_25),demonstrating substantial improvements over conventional SSL approaches.Our results indicate that the synergistic integration of LLMs,diffusion models,and the Swin Transformer architecture contributes significantly to recognition accuracy,particularly for rare and morphologically similar species.These findings highlight the potential of SDNet for addressing fundamental limitations of existing SSL methods in avian recognition tasks and establishing a new paradigm for efficient self-supervised learning in large-scale ornithological vision applications.
基金supported in part by the National Natural Science Foundation of China under Grant 62071405the National Natural Science Foundation of China under Grant 12175189.
文摘The accurate segmentation of deep gray matter nuclei is critical for neuropathological research,disease diagnosis and treatment.Existing methods employ the supervised learning training approach,which requires large labeled datasets.It is challenging and time-consuming to obtain such datasets for medical image analysis.In addition,these methods based on convolutional neural networks(CNNs)only achieve suboptimal performance due to the locality of convolutional operations.Vision Transformers(ViTs)efficiently model long-range dependencies and thus have the potentiality to outperform these methods in segmentation tasks.To address these issues,we propose a novel hybrid network based on self-supervised pre-training for deep gray matter nuclei segmentation.Specifically,we present a CNN-Transformer hybrid network(CTNet),whose encoder consists of 3D CNN and ViT to learn local spatial-detailed features and global semantic information.A self-supervised learning(SSL)approach that integrates rotation prediction and masked feature reconstruction is proposed to pre-train the CTNet,enabling the model to learn valuable visual representations from unlabeled data.We evaluate the effectiveness of our method on 3T and 7T human brain MRI datasets.The results demonstrate that our CTNet achieves better performance than other comparison models and our pre-training strategy outperforms other advanced self-supervised methods.When the training set has only one sample,our pre-trained CTNet enhances segmentation performance,showing an 8.4%improvement in Dice similarity coefficient(DSC)compared to the randomly initialized CTNet.
基金supported by the National Key Research and Development Program of China(Program No.:2022YFF1203003)the National Natural Science Foundation of China(Grant No.:82373791).
文摘P-glycoprotein(P-gp)is a transmembrane protein widely involved in the absorption,distribution,metabolism,excretion,and toxicity(ADMET)of drugs within the human body.Accurate prediction of Pgp inhibitors and substrates is crucial for drug discovery and toxicological assessment.However,existing models rely on limited molecular information,leading to suboptimal model performance for predicting P-gp inhibitors and substrates.To overcome this challenge,we compiled an extensive dataset from public databases and literature,consisting of 5,943 P-gp inhibitors and 4,018 substrates,notable for their high quantity,quality,and structural uniqueness.In addition,we curated two external test sets to validate the model's generalization capability.Subsequently,we developed a multimodal graph contrastive learning(GCL)model for the prediction of P-gp inhibitors and substrates(MC-PGP).This framework integrates three types of features from Simplified Molecular Input Line Entry System(SMILES)sequences,molecular fingerprints,and molecular graphs using an attention-based fusion strategy to generate a unified molecular representation.Furthermore,we employed a GCL approach to enhance structural representations by aligning local and global structures.Extensive experimental results highlight the superior performance of MC-PGP,which achieves improvements in the area under the curve of receiver operating characteristic(AUC-ROC)of 9.82%and 10.62%on the external P-gp inhibitor and external P-gp substrate datasets,respectively,compared with 12 state-of-the-art methods.Furthermore,the interpretability analysis of all three molecular feature types offers comprehensive and complementary insights,demonstrating that MC-PGP effectively identifies key functional groups involved in P-gp interactions.These chemically intuitive insights provide valuable guidance for the design and optimization of drug candidates.
基金supported by the ScientificResearch and Innovation Team Program of Sichuan University of Science and Technology(No.SUSE652A006)Sichuan Key Provincial Research Base of Intelligent Tourism(ZHYJ22-03)In addition,it is also listed as a project of Sichuan Provincial Science and Technology Programme(2022YFG0028).
文摘Emotion recognition plays a crucial role in various fields and is a key task in natural language processing (NLP). The objective is to identify and interpret emotional expressions in text. However, traditional emotion recognition approaches often struggle in few-shot cross-domain scenarios due to their limited capacity to generalize semantic features across different domains. Additionally, these methods face challenges in accurately capturing complex emotional states, particularly those that are subtle or implicit. To overcome these limitations, we introduce a novel approach called Dual-Task Contrastive Meta-Learning (DTCML). This method combines meta-learning and contrastive learning to improve emotion recognition. Meta-learning enhances the model’s ability to generalize to new emotional tasks, while instance contrastive learning further refines the model by distinguishing unique features within each category, enabling it to better differentiate complex emotional expressions. Prototype contrastive learning, in turn, helps the model address the semantic complexity of emotions across different domains, enabling the model to learn fine-grained emotions expression. By leveraging dual tasks, DTCML learns from two domains simultaneously, the model is encouraged to learn more diverse and generalizable emotions features, thereby improving its cross-domain adaptability and robustness, and enhancing its generalization ability. We evaluated the performance of DTCML across four cross-domain settings, and the results show that our method outperforms the best baseline by 5.88%, 12.04%, 8.49%, and 8.40% in terms of accuracy.
基金supported by the Natural Science Foundation of Xinjiang Uygur Autonomous Region(Grant No.2022D01B 187)。
文摘Federated learning(FL)is a distributed machine learning paradigm for edge cloud computing.FL can facilitate data-driven decision-making in tactical scenarios,effectively addressing both data volume and infrastructure challenges in edge environments.However,the diversity of clients in edge cloud computing presents significant challenges for FL.Personalized federated learning(pFL)received considerable attention in recent years.One example of pFL involves exploiting the global and local information in the local model.Current pFL algorithms experience limitations such as slow convergence speed,catastrophic forgetting,and poor performance in complex tasks,which still have significant shortcomings compared to the centralized learning.To achieve high pFL performance,we propose FedCLCC:Federated Contrastive Learning and Conditional Computing.The core of FedCLCC is the use of contrastive learning and conditional computing.Contrastive learning determines the feature representation similarity to adjust the local model.Conditional computing separates the global and local information and feeds it to their corresponding heads for global and local handling.Our comprehensive experiments demonstrate that FedCLCC outperforms other state-of-the-art FL algorithms.
文摘Graph similarity learning aims to calculate the similarity between pairs of graphs.Existing unsupervised graph similarity learning methods based on contrastive learning encounter challenges related to random graph augmentation strategies,which can harm the semantic and structural information of graphs and overlook the rich structural information present in subgraphs.To address these issues,we propose a graph similarity learning model based on learnable augmentation and multi-level contrastive learning.First,to tackle the problem of random augmentation disrupting the semantics and structure of the graph,we design a learnable augmentation method to selectively choose nodes and edges within the graph.To enhance contrastive levels,we employ a biased random walk method to generate corresponding subgraphs,enriching the contrastive hierarchy.Second,to solve the issue of previous work not considering multi-level contrastive learning,we utilize graph convolutional networks to learn node representations of augmented views and the original graph and calculate the interaction information between the attribute-augmented and structure-augmented views and the original graph.The goal is to maximize node consistency between different views and learn node matching between different graphs,resulting in node-level representations for each graph.Subgraph representations are then obtained through pooling operations,and we conduct contrastive learning utilizing both node and subgraph representations.Finally,the graph similarity score is computed according to different downstream tasks.We conducted three sets of experiments across eight datasets,and the results demonstrate that the proposed model effectively mitigates the issues of random augmentation damaging the original graph’s semantics and structure,as well as the insufficiency of contrastive levels.Additionally,the model achieves the best overall performance.
基金supported by the Joint Funds of the National Natural Science Foundation of China under grant No.U22B2025.
文摘In the big data era,the surge in network traffic volume poses challenges for network management and cybersecurity.Network Traffic Classification(NTC)employs deep learning to categorize traffic data,aiding security and analysis systems as Intrusion Detection Systems(IDS)and Intrusion Prevention Systems(IPS).However,current NTC methods,based on isolated network simulations,usually fail to adapt to new protocols and applications and ignore the effects of network conditions and user behavior on traffic patterns.To improve network traffic management insights,federated learning frameworks have been proposed to aggregate diverse traffic data for collaborative model training.This approach faces challenges like data integrity,label noise,packet loss,and skewed data distributions.While label noise can be mitigated through the use of sophisticated traffic labeling tools,other issues such as packet loss and skewed data distributions encountered in Network Packet Brokers(NPB)can severely impede the efficacy of federated learning algorithms.In this paper,we introduced the Robust Traffic Classifier with Federated Contrastive Learning(FC-RTC),combining federated and contrastive learning methods.Using the Supcon-Loss function from contrastive learning,FC-RTC distinguishes between similar and dissimilar samples.Training by sample pairs,FC-RTC effectively updates when receiving corrupted traffic data with packet loss or disorder.In cases of sample imbalance,contrastive loss functions for similar samples reduce model bias towards higher proportion data.By addressing uneven data distribution and packet loss,our system enhances its capability to adapt and perform accurately in real-world network traffic analysis,meeting the specific demands of this complex field.
文摘Fisheye cameras offer a significantly larger field of view compared to conventional cameras,making them valuable tools in the field of computer vision.However,their unique optical characteristics often lead to image distortions,which pose challenges for object detection tasks.To address this issue,we propose Yolo-CaSKA(Yolo with Contrastive Learning and Selective Kernel Attention),a novel training method that enhances object detection on fisheye camera images.The standard image and the corresponding distorted fisheye image pairs are used as positive samples,and the rest of the image pairs are used as negative samples,which are guided by contrastive learning to help the distorted images find the feature vectors of the corresponding normal images,to improve the detection accuracy.Additionally,we incorporate the Selective Kernel(SK)attention module to focus on regions prone to false detections,such as image edges and blind spots.Finally,the mAP_(50) on the augmented KITTI dataset is improved by 5.5% over the original Yolov8,while the mAP_(50) on the WoodScape dataset is improved by 2.6% compared to OmniDet.The results demonstrate the performance of our proposed model for object detection on fisheye images.
基金funded by the China Chongqing Municipal Science and Technology Bureau,grant numbers CSTB2024TIAD-CYKJCXX0009,CSTB2024NSCQ-LZX0043,CSTB2022NSCQ-MSX0288Chongqing Municipal Commission of Housing and Urban-Rural Development,grant number CKZ2024-87+3 种基金the Chongqing University of Technology Graduate Education High-Quality Development Project,grant number gzlsz202401the Chongqing University of Technology—Chongqing LINGLUE Technology Co.,Ltd.Electronic Information(Artificial Intelligence)Graduate Joint Training Basethe Postgraduate Education and Teaching Reform Research Project in Chongqing,grant number yjg213116the Chongqing University of Technology-CISDI Chongqing Information Technology Co.,Ltd.Computer Technology Graduate Joint Training Base.
文摘Although conventional object detection methods achieve high accuracy through extensively annotated datasets,acquiring such large-scale labeled data remains challenging and cost-prohibitive in numerous real-world applications.Few-shot object detection presents a new research idea that aims to localize and classify objects in images using only limited annotated examples.However,the inherent challenge in few-shot object detection lies in the insufficient sample diversity to fully characterize the sample feature distribution,which consequently impacts model performance.Inspired by contrastive learning principles,we propose an Implicit Feature Contrastive Learning(IFCL)module to address this limitation and augment feature diversity for more robust representational learning.This module generates augmented support sample features in a mixed feature space and implicitly contrasts them with query Region of Interest(RoI)features.This approach facilitates more comprehensive learning of both intra-class feature similarity and inter-class feature diversity,thereby enhancing the model’s object classification and localization capabilities.Extensive experiments on PASCAL VOC show that our method achieves a respective improvement of 3.2%,1.8%,and 2.3%on 10-shot of three Novel Sets compared to the baseline model FPD.
基金granted by Qin Xin Talents Cultivation Program(No.QXTCP C202115),Beijing Information Science&Technology Universitythe Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing Fund(No.GJJ-23),National Social Science Foundation,China(No.21BTQ079).
文摘Sarcasm detection is a complex and challenging task,particularly in the context of Chinese social media,where it exhibits strong contextual dependencies and cultural specificity.To address the limitations of existing methods in capturing the implicit semantics and contextual associations in sarcastic expressions,this paper proposes an event-aware model for Chinese sarcasm detection,leveraging a multi-head attention(MHA)mechanism and contrastive learning(CL)strategies.The proposed model employs a dual-path Bidirectional Encoder Representations from Transformers(BERT)encoder to process comment text and event context separately and integrates an MHA mechanism to facilitate deep interactions between the two,thereby capturing multidimensional semantic associations.Additionally,a CL strategy is introduced to enhance feature representation capabilities,further improving the model’s performance in handling class imbalance and complex contextual scenarios.The model achieves state-of-the-art performance on the Chinese sarcasm dataset,with significant improvements in accuracy(79.55%),F1-score(84.22%),and an area under the curve(AUC,84.35%).
文摘In the field of optoelectronics,certain types of data may be difficult to accurately annotate,such as high-resolution optoelectronic imaging or imaging in certain special spectral ranges.Weakly supervised learning can provide a more reliable approach in these situations.Current popular approaches mainly adopt the classification-based class activation maps(CAM)as initial pseudo labels to solve the task.
基金supported in part by the National Natural Science Foundation of China(62472032)the Young Elite Scientists Sponsorship Program by CAST(2023QNRC001)the Fundamental Research Funds for the Central Universities and Research Innovation Project of China University of Political Science and Law(21ZFY52001)。
文摘In federated learning(FL),the distribution of data across different clients leads to the degradation of global model performance in training.Personalized Federated Learning(pFL)can address this problem through global model personalization.Researches over the past few years have calibrated differences in weights across the entire model or optimized only individual layers of the model without considering that different layers of the whole neural network have different utilities,resulting in lagged model convergence and inadequate personalization in non-IID data.In this paper,we propose model layered optimization for feature extractor and classifier(pFedEC),a novel pFL training framework personalized for different layers of the model.Our study divides the model layers into the feature extractor and classifier.We initialize the model's classifiers during model training,while making the local model's feature extractors learn the representation of the global model's feature extractors to correct each client's local training,integrating the utilities of the different layers in the entire model.Our extensive experiments show that pFedEC achieves 92.95%accuracy on CIFAR-10,outperforming existing pFL methods by approximately 1.8%.On CIFAR-100 and Tiny-ImageNet,pFedEC improves the accuracy by at least 4.2%,reaching 73.02%and 28.39%,respectively.
文摘Few-shot point cloud 3D object detection(FS3D)aims to identify and locate objects of novel classes within point clouds using knowledge acquired from annotated base classes and a minimal number of samples from the novel classes.Due to imbalanced training data,existing FS3D methods based on fully supervised learning can lead to overfitting toward base classes,which impairs the network’s ability to generalize knowledge learned from base classes to novel classes and also prevents the network from extracting distinctive foreground and background representations for novel class objects.To address these issues,this thesis proposes a category-agnostic contrastive learning approach,enhancing the generalization and identification abilities for almost unseen categories through the construction of pseudo-labels and positive-negative sample pairs unrelated to specific classes.Firstly,this thesis designs a proposal-wise context contrastive module(CCM).By reducing the distance between foreground point features and increasing the distance between foreground and background point features within a region proposal,CCM aids the network in extracting more discriminative foreground and background feature representations without reliance on categorical annotations.Secondly,this thesis utilizes a geometric contrastive module(GCM),which enhances the network’s geometric perception capability by employing contrastive learning on the foreground point features associated with various basic geometric components,such as edges,corners,and surfaces,thereby enabling these geometric components to exhibit more distinguishable representations.This thesis also combines category-aware contrastive learning with former modules to maintain categorical distinctiveness.Extensive experimental results on FS-SUNRGBD and FS-ScanNet datasets demonstrate the effectiveness of this method with average precision exceeding the baseline by up to 8%.
