Accurate and reliable fault diagnosis is critical for secure operation in complex smart power systems.While graph neural networks show promise for this task,existing methods often neglect the long-tailed distribution ...Accurate and reliable fault diagnosis is critical for secure operation in complex smart power systems.While graph neural networks show promise for this task,existing methods often neglect the long-tailed distribution inherent in real-world grid fault data and fail to provide reliability estimates for their decisions.To address these dual challenges,we propose a novel multi-expert collaboration uncertainty-aware power fault recognition framework with cross-view graph learning.Its core innovations are two synergistic modules:(1)The infographics aggregation module tackles the long-tail problem by learning robust graph-level representations.It employs an information-driven optimization loss within a contrastive graph architecture,explicitly preserving global invariance and local structural information across diverse(including rare)fault states.This ensures balanced representation learning for both the head and tail classes.(2)The multi-expert reliable decision module addresses prediction uncertainty.It trains individual expert classifiers using the Dirichlet distribution to explicitly model the credibility(uncertainty)of each expert’s decision.Crucially,a complementary collaboration rule based on evidence theory dynamically integrates these experts.This rule generates active weights for expert participation,prioritizing more certain experts and fusing their evidence to produce a final decision with a quantifiable reliability estimate.Collaboratively,these modules enable reliable diagnosis under data imbalance:The Infographics Module provides discriminative representations for all fault types,especially tail classes,while the Multi-Expert Module leverages these representations to make decisions with explicit uncertainty quantification.This synergy significantly improves both the accuracy and the reliability of fault recognition,particularly for rare or ambiguous grid conditions.Ultimately,extensive experiment evaluations on the four datasets reveal that the proposed method outperforms the state-of-the-art methods in the fault diagnosis of smart grids,in terms of accuracy,precision,f score and recall.展开更多
Underwater image enhancement aims to restore a clean appearance and thus improves the quality of underwater degraded images.Current methods feed the whole image directly into the model for enhancement.However,they ign...Underwater image enhancement aims to restore a clean appearance and thus improves the quality of underwater degraded images.Current methods feed the whole image directly into the model for enhancement.However,they ignored that the R,G and B channels of underwater degraded images present varied degrees of degradation,due to the selective absorption for the light.To address this issue,we propose an unsupervised multi-expert learning model by considering the enhancement of each color channel.Specifically,an unsupervised architecture based on generative adversarial network is employed to alleviate the need for paired underwater images.Based on this,we design a generator,including a multi-expert encoder,a feature fusion module and a feature fusion-guided decoder,to generate the clear underwater image.Accordingly,a multi-expert discriminator is proposed to verify the authenticity of the R,G and B channels,respectively.In addition,content perceptual loss and edge loss are introduced into the loss function to further improve the content and details of the enhanced images.Extensive experiments on public datasets demonstrate that our method achieves more pleasing results in vision quality.Various metrics(PSNR,SSIM,UIQM and UCIQE) evaluated on our enhanced images have been improved obviously.展开更多
针对现有的专家推荐算法忽略了用户评论中蕴含的情感表达对专家专长表征的影响,从而导致推荐准确度不高的问题,提出基于双向编码器表示-多头注意力机制(bidirectional encoder representations from transformers-multi-head attention,...针对现有的专家推荐算法忽略了用户评论中蕴含的情感表达对专家专长表征的影响,从而导致推荐准确度不高的问题,提出基于双向编码器表示-多头注意力机制(bidirectional encoder representations from transformers-multi-head attention,BERT-MHA)的深度语义增强专家推荐算法。该算法基于预训练BERT模型,融合MHA机制,自动调整用户评论对专家历史回答问题的情感注意力权重,获取专家动态专长表征,并与静态专长联合以实现专家特征文本的语义增强,表征专家综合专长;通过注意力机制识别用户问题特征;采用多层感知机建模专家综合专长与用户问题间的非线性交互,预测推荐专家的匹配度。利用好大夫网站(haodf.com)的数据进行了不同参数配置及不同算法的对比实验,实验结果表明该算法在准确率(accuracy,ACC)和曲线下的面积(area under curve,AUC)指标下明显优于其他算法,能有效提高在线问答社区的专家推荐准确度。展开更多
近年来,大语言模型(large language models,LLMs)在自然语言处理等领域展现出卓越性能。然而,在显存受限的工业环境中,将LLMs扩展至多个下游任务时常难以兼顾资源消耗与性能表现之间的平衡,进而限制了其更广泛的应用。为此,提出了一种...近年来,大语言模型(large language models,LLMs)在自然语言处理等领域展现出卓越性能。然而,在显存受限的工业环境中,将LLMs扩展至多个下游任务时常难以兼顾资源消耗与性能表现之间的平衡,进而限制了其更广泛的应用。为此,提出了一种结构紧凑的多专家协同架构(compact llm with collaboration of experts,CCoE)。该架构采用模块化设计,能够高效且灵活地将多个领域专家集成至统一的LLMs中,在保证性能的同时显著降低多专家部署的显存开销。此外,CCoE通过引人基于规则的门控机制与专家规划模块,实现了任务的精准分配与专家间的协同,从而有效支持复杂推理任务。在5个领域数据集上的实验证明,CCoE在各项任务中均达到了与现有垂类LLMs相当的表现。此外,相比于现有模型集成方法,CCoE在保持性能的前提下将显存占用量降低了61.3%,并较参数高效的多专家集成方法提升了76.4%的推理吞吐量。展开更多
基金supported by the Development Department Science and Technology Project(52992624000X).
文摘Accurate and reliable fault diagnosis is critical for secure operation in complex smart power systems.While graph neural networks show promise for this task,existing methods often neglect the long-tailed distribution inherent in real-world grid fault data and fail to provide reliability estimates for their decisions.To address these dual challenges,we propose a novel multi-expert collaboration uncertainty-aware power fault recognition framework with cross-view graph learning.Its core innovations are two synergistic modules:(1)The infographics aggregation module tackles the long-tail problem by learning robust graph-level representations.It employs an information-driven optimization loss within a contrastive graph architecture,explicitly preserving global invariance and local structural information across diverse(including rare)fault states.This ensures balanced representation learning for both the head and tail classes.(2)The multi-expert reliable decision module addresses prediction uncertainty.It trains individual expert classifiers using the Dirichlet distribution to explicitly model the credibility(uncertainty)of each expert’s decision.Crucially,a complementary collaboration rule based on evidence theory dynamically integrates these experts.This rule generates active weights for expert participation,prioritizing more certain experts and fusing their evidence to produce a final decision with a quantifiable reliability estimate.Collaboratively,these modules enable reliable diagnosis under data imbalance:The Infographics Module provides discriminative representations for all fault types,especially tail classes,while the Multi-Expert Module leverages these representations to make decisions with explicit uncertainty quantification.This synergy significantly improves both the accuracy and the reliability of fault recognition,particularly for rare or ambiguous grid conditions.Ultimately,extensive experiment evaluations on the four datasets reveal that the proposed method outperforms the state-of-the-art methods in the fault diagnosis of smart grids,in terms of accuracy,precision,f score and recall.
基金supported in part by the National Key Research and Development Program of China(2020YFB1313002)the National Natural Science Foundation of China(62276023,U22B2055,62222302,U2013202)+1 种基金the Fundamental Research Funds for the Central Universities(FRF-TP-22-003C1)the Postgraduate Education Reform Project of Henan Province(2021SJGLX260Y)。
文摘Underwater image enhancement aims to restore a clean appearance and thus improves the quality of underwater degraded images.Current methods feed the whole image directly into the model for enhancement.However,they ignored that the R,G and B channels of underwater degraded images present varied degrees of degradation,due to the selective absorption for the light.To address this issue,we propose an unsupervised multi-expert learning model by considering the enhancement of each color channel.Specifically,an unsupervised architecture based on generative adversarial network is employed to alleviate the need for paired underwater images.Based on this,we design a generator,including a multi-expert encoder,a feature fusion module and a feature fusion-guided decoder,to generate the clear underwater image.Accordingly,a multi-expert discriminator is proposed to verify the authenticity of the R,G and B channels,respectively.In addition,content perceptual loss and edge loss are introduced into the loss function to further improve the content and details of the enhanced images.Extensive experiments on public datasets demonstrate that our method achieves more pleasing results in vision quality.Various metrics(PSNR,SSIM,UIQM and UCIQE) evaluated on our enhanced images have been improved obviously.
文摘针对现有的专家推荐算法忽略了用户评论中蕴含的情感表达对专家专长表征的影响,从而导致推荐准确度不高的问题,提出基于双向编码器表示-多头注意力机制(bidirectional encoder representations from transformers-multi-head attention,BERT-MHA)的深度语义增强专家推荐算法。该算法基于预训练BERT模型,融合MHA机制,自动调整用户评论对专家历史回答问题的情感注意力权重,获取专家动态专长表征,并与静态专长联合以实现专家特征文本的语义增强,表征专家综合专长;通过注意力机制识别用户问题特征;采用多层感知机建模专家综合专长与用户问题间的非线性交互,预测推荐专家的匹配度。利用好大夫网站(haodf.com)的数据进行了不同参数配置及不同算法的对比实验,实验结果表明该算法在准确率(accuracy,ACC)和曲线下的面积(area under curve,AUC)指标下明显优于其他算法,能有效提高在线问答社区的专家推荐准确度。
文摘近年来,大语言模型(large language models,LLMs)在自然语言处理等领域展现出卓越性能。然而,在显存受限的工业环境中,将LLMs扩展至多个下游任务时常难以兼顾资源消耗与性能表现之间的平衡,进而限制了其更广泛的应用。为此,提出了一种结构紧凑的多专家协同架构(compact llm with collaboration of experts,CCoE)。该架构采用模块化设计,能够高效且灵活地将多个领域专家集成至统一的LLMs中,在保证性能的同时显著降低多专家部署的显存开销。此外,CCoE通过引人基于规则的门控机制与专家规划模块,实现了任务的精准分配与专家间的协同,从而有效支持复杂推理任务。在5个领域数据集上的实验证明,CCoE在各项任务中均达到了与现有垂类LLMs相当的表现。此外,相比于现有模型集成方法,CCoE在保持性能的前提下将显存占用量降低了61.3%,并较参数高效的多专家集成方法提升了76.4%的推理吞吐量。
