A5101735689- Body Composition Measurement Techniques
- Anomaly Detection Techniques and Applications
- Domain Adaptation and Few-Shot Learning
- Video Surveillance and Tracking Methods
- Mobile Crowdsensing and Crowdsourcing
- Heat Transfer and Optimization
- Machine Learning and Data Classification
- Thermal properties of materials
- Neural Networks and Applications
- Advanced Neural Network Applications
- Data Stream Mining Techniques
- Music and Audio Processing
- Privacy-Preserving Technologies in Data
- Neural Networks and Reservoir Computing
- Fire Detection and Safety Systems
- Advanced Clustering Algorithms Research
- Thermal Radiation and Cooling Technologies
- Advanced Image and Video Retrieval Techniques
- Adversarial Robustness in Machine Learning
- Second Language Learning and Teaching
- Education and Critical Thinking Development
- Generative Adversarial Networks and Image Synthesis
- Antimicrobial Peptides and Activities
- Antimicrobial agents and applications
- Medicinal Plant Pharmacodynamics Research
Institute of Information Engineering
2019-2024
Chinese Academy of Sciences
2019-2024
University of Chinese Academy of Sciences
2019-2024
Institute of Engineering Thermophysics
2021-2023
Nanjing Institute of Technology
2023
North China Electric Power University
2023
Qingdao University of Science and Technology
2022
Nanjing Normal University
2020-2021
University of Exeter
2019
Beijing University of Posts and Telecommunications
2015
Deep networks have strong capacities of embedding data into latent representations and finishing following tasks. However, the largely come from high-quality annotated labels, which are expensive to collect. Noisy labels more affordable, but result in corrupted representations, leading poor generalization performance. To learn robust handle noisy we propose selective-supervised contrastive learning (Sel-CL) this paper. Specifically, Sel-CL extend supervised (Sup-CL), is powerful...
Knowledge distillation provides an effective way to transfer knowledge via teacher-student learning, where most existing approaches apply a fixed pre-trained model as teacher supervise the learning of student network. This manner usually brings in big capability gap between and networks during learning. Recent researches have observed that small can facilitate transfer. Inspired by that, we propose evolutionary approach improve effectiveness knowledge. Instead teacher, is learned online...
The Vision Meets Drone (VisDrone2019) Single Object Tracking challenge is the second annual research activity focusing on evaluating single-object tracking algorithms drones, held in conjunction with International Conference Computer (ICCV 2019). VisDrone-SOT2019 Challenge goes beyond its VisDrone-SOT2018 predecessor by introducing 25 more challenging sequences for long-term tracking. We evaluate and discuss results of 22 participating 19 state-of-the-art trackers collected dataset. are...
Learning from crowds describes that the annotations of training data are obtained with crowd-sourcing services. Multiple annotators each complete their own small part annotations, where labeling mistakes depend on occur frequently. Modeling label-noise generation process by noise transition matrix is a powerful tool to tackle label noise. In real-world scenarios, matrices both annotator- and instance-dependent. However, due high complexity instance-dependent (AIDTM), <italic...
Microbial infections have become a great threat to human health and one of the main risks arises from direct contact with surfaces contaminated by pathogenic microbes. Herein, kind hexagonal column interpenetrated spheres (HCISs) are fabricated non-covalent assembly plant gallic acid quaternary ammonium surfactants. Different one-time burst release conventional antimicrobial agents, HCIS acts like "antimicrobial molecular bank" releases ingredients in multistage way, leading long-lasting...
As the size of datasets getting larger, accurately annotating such is becoming more impractical due to expensiveness on both time and economy. Therefore, crowd-sourcing has been widely adopted alleviate cost collecting labels, which also inevitably introduces label noise eventually degrades performance model. To learn from annotations, modeling expertise each annotator a common but challenging paradigm, because annotations collected by are usually highly-sparse. this problem, we propose...
Training state-of-the-art (SOTA) deep models often requires extensive data, resulting in substantial training and storage costs. To address these challenges, dataset condensation has been developed to learn a small synthetic set that preserves essential information from the original large-scale dataset. Nowadays, optimization-oriented methods have primary method field of for achieving SOTA results. However, bi-level optimization process hinders practical application such realistic larger...
Typically, learning a deep classifier from massive cleanly annotated instances is effective but impractical in many real-world scenarios. An alternative collecting and aggregating multiple noisy annotations for each instance to train the classifier. Inspired by that, this paper proposes learn annotators via coupled-view approach, where view data represented neural networks classification labels described Naive Bayes label aggregation. Such converted supervised problem under mutual...
Recent deep trackers have shown superior performance in visual tracking. In this article, we propose a cascaded correlation refinement approach to facilitate the robustness of The core idea is address accurate target localization and reliable model update collaborative way. To end, our cascades multiple stages progressively refine localization. Thus, localized object could be used learn an on-the-fly for improving reliability update. Meanwhile, introduce explicit measure identify tracking...
Heat accumulation generated from confined space poses a threat to the service reliability and lifetime of electronic devices. To quickly remove excess heat hot spot, it is highly desirable enhance dissipation in specific direction. Herein, we report facile route fabricate large-scale composite film with enhanced thermal conductivity electrical insulation. The well-stacked films were constructed by assembly polydopamine (PDA)-modified graphene nanosheets (GNSPDA) hexagonal boron nitride...
Composed query image retrieval task aims to retrieve the target in database by a that composes two different modalities: reference and sentence declaring some details of need be modified replaced new elements. Tackling this needs learn multimodal embedding space, which can make semantically similar targets queries close but dissimilar as far away possible. Most existing methods start from perspective model structure design clever interactive modules promote better fusion modalities. However,...
Supervised deep learning depends on massive accurately annotated examples, which is usually impractical in many real-world scenarios. A typical alternative from multiple noisy annotators. Numerous earlier works assume that all labels are noisy, while it the case a few trusted samples with clean available. This raises following important question: how can we effectively use small amount of data to facilitate robust classifier annotators? paper proposes data-efficient approach, called...
Federated learning provides a privacy-preserving manner to collaboratively train models on data distributed over multiple local clients via the coordination of global server. In this paper, we focus label distribution skew in federated learning, where due different user behavior client, distributions between are significantly different. When faced with such cases, most existing methods will lead suboptimal optimization inadequate utilization information clients. Inspired by this, propose...
Multilingual teaching and learning practices are often implemented in K-12 classrooms. However, issues related to multilingual assessment rarely investigated. With the growing population of learners classroom, there is a great need understand what how assess students who come from culturally linguistically diverse backgrounds. The current study attempts fill research gap by reviewing literature over past 15 years on multilingualism. We summarize synthesize three main themes: 1)...
The thermal performance of the microgrooved heat pipes is mainly restricted by wick capillary limit. construction micro-nano structures and using nanofluids are effective ways to improve microgroove wetting characteristic transfer ability. Based on accommodation theory, a model for nanofluid characteristics in pipe was presented this paper. axial ability SiO2 untreated rectangular microgrooves (RMs) with superhydrophilic nano-textured surfaces (RMSNSs) were compared. results indicate that...
Hierarchical control architecture is designed for software-defined multidomain optical networks (SD-MDONs), and a unified service logic processing model (USLPM) first proposed various applications. USLPM-based virtual network (VON) provisioning process designed, two VON mapping algorithms are proposed: random node selection per controller computation (RNS&PCC) balanced hierarchical (BNS&HCC). Then an SD-MDON testbed built with OpenFlow extension in order to support transport equipment....
Low-resolution face recognition in the wild still is an open problem. In this paper, we propose to address problem via a novel learning approach called Mixed-Domain Distillation (MDD). The applies teacher-student framework mix and distill knowledge from four different domain datasets, including private high-resolution, public low-resolution web target datasets. way, high-resolution well-trained complex teacher model first adapted faces then transferred simply student model. designed identify...
While massive valuable deep models trained on large-scale data have been released to facilitate the artificial intelligence community, they may encounter attacks in deployment which leads privacy leakage of training data. In this work, we propose a learning approach termed differentially private data-free distillation (DPDFD) for model conversion that can convert pretrained (teacher) into its privacy-preserving counterpart (student) via an intermediate generator without access The...
Abstract Microbial infections have become a great threat to human health and one of the main risks arises from direct contact with surfaces contaminated by pathogenic microbes. Herein, kind hexagonal column interpenetrated spheres (HCISs) are fabricated non‐covalent assembly plant gallic acid quaternary ammonium surfactants. Different one‐time burst release conventional antimicrobial agents, HCIS acts like “antimicrobial molecular bank” releases ingredients in multistage way, leading...
Learning from crowds describes that the annotations of training data are obtained with crowd-sourcing services. Multiple annotators each complete their own small part annotations, where labeling mistakes depend on occur frequently. Modeling label-noise generation process by noise transition matrix is a power tool to tackle label noise. In real-world scenarios, matrices both annotator- and instance-dependent. However, due high complexity instance-dependent (AIDTM), annotation sparsity, which...
Dataset condensation addresses the problem of data burden by learning a small synthetic training set that preserves essential knowledge from larger real set. To date, state-of-the-art (SOTA) results are often yielded optimization-oriented methods, but their inefficiency hinders application to realistic datasets. On other hand, Distribution-Matching (DM) methods show remarkable efficiency sub-optimal compared methods. In this paper, we reveal limitations current DM-based inner-class and...
Dataset condensation addresses the problem of data burden by learning a small synthetic training set that preserves essential knowledge from larger real set. To date, state-of-the-art (SOTA) results are often yielded optimization-oriented methods, but their inefficiency hinders application to realistic datasets. On other hand, Distribution-Matching (DM) methods show remarkable efficiency sub-optimal compared methods. In this paper, we reveal limitations current DM-based inner-class and...