A5100439506- Anomaly Detection Techniques and Applications
- Domain Adaptation and Few-Shot Learning
- Video Surveillance and Tracking Methods
- Face and Expression Recognition
- Advanced Neural Network Applications
- Network Security and Intrusion Detection
- Machine Learning and Data Classification
- Face recognition and analysis
- Remote-Sensing Image Classification
- Advanced Graph Neural Networks
- Machine Learning and ELM
- Fault Detection and Control Systems
- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Human Pose and Action Recognition
- Metaheuristic Optimization Algorithms Research
- Complex Network Analysis Techniques
- Data Quality and Management
- Data Stream Mining Techniques
- Multimodal Machine Learning Applications
- Topic Modeling
- Quantum Information and Cryptography
- Video Analysis and Summarization
- Text and Document Classification Technologies
- Artificial Intelligence Applications
Henan Normal University
2024
Inner Mongolia Academy of Agricultural & Animal Husbandry Sciences
2024
BGI Group (China)
2024
BGI Genomics
2024
Southern Methodist University
2024
Zhejiang Lab
2021-2023
Suzhou Institute of Biomedical Engineering and Technology
2023
Chinese Academy of Sciences
2023
Taiyuan University of Technology
2023
Xidian University
2019-2023
The goal of semi-supervised object detection is to learn a model using only few labeled data and large amounts unlabeled data, thereby reducing the cost labeling. Although studies have proposed various self-training-based methods or consistency regularization-based methods, they ignore discrepancies among results in same image that occur during different training iterations. Additionally, predicted vary models. In this paper, we propose an interactive form self-training mean teachers for...
In crowd counting, due to the problem of laborious labelling, it is perceived intractability collecting a new large-scale dataset which has plentiful images with large diversity in density, scene, etc. Thus, for learning general model, training data from multiple different datasets might be remedy and great value. this paper, we resort multi-domain joint propose simple but effective Domain-specific Knowledge Propagating Network (DKPNet) unbiasedly knowledge diverse domains at same time. It...
Recently, deep learning-based crowd counting methods have achieved promising performance on test data with the same distribution as training set, while degradation usually occurs when testing other or unseen domains. Due to variations in scene contexts, densities and head scales, it is a very challenging issue tackle multi-domain using one model. In this work, we propose domain-guided channel attention network (DCANet) towards learning counting. particular, our DCANet consists of feature...
Advancements in sequencing have enabled the assembly of numerous sheep genomes, significantly advancing our understanding link between genetic variation and phenotypic traits. However, genome East Friesian (Ostfriesisches Milchschaf), a key high-yield milk breed, remains to be fully assembled. Here, we constructed near-complete gap-free using PacBio HiFi, ultra-long ONT Hi-C sequencing. The resulting spans approximately 2.96 Gb, with contig N50 length 104.1 Mb only 164 unplaced sequences....
Earlier-stage evaluations of a new AI architecture/system need affordable benchmarks. Only using few component benchmarks like MLPerf alone in the other stages may lead to misleading conclusions. Moreover, learning dynamics are not well understood, and benchmarks' shelf-life is short. This paper proposes balanced benchmarking methodology. We use real-world cover factors space that impacts most considerable extent. After performing an exhaustive survey on Internet service domains, we identify...
The goal of video highlight detection is to select the most attractive segments from a long depict interesting parts video. Existing methods typically focus on modeling relationship between different in order learning model that can assign scores these segments; however, approaches do not explicitly consider contextual dependency within individual segments. To this end, we propose learn pixel-level distinctions improve detection. This distinction indicates whether or each pixel one belongs...
This paper focuses on the issue of anomaly detection for data in process control systems (PCSs). Considering features PCSs, this proposes to utilize notion one-class classification (OCC). In order provide a general solution more types systems, ensemble learning is combined with OCC models. Two different ensembles models are proposed based scenarios detection. Performance scheme validated via several UCI datasets and two practical PCSs.
Recently, deep face recognition has achieved significant progress because of Convolutional Neural Networks (CNNs) and large-scale datasets. However, training CNNs on a dataset with limited computational resources is still challenge. This the classification paradigm needs to train fully-connected layer as category classifier, its parameters will be in hundreds millions if contains identities. requires many resources, such GPU memory. The metric learning an economical computation method, but...
In many applications of the analysis dynamic graph, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Timing iterative Graph Processing</i> (TGP) jobs usually need to be generated for processing corresponding snapshots graph obtain results at different points time. For high throughput such applications, it is expected run TGP on GPU concurrently. Although GPU-based systems have been recently developed, out-of-GPU-memory processing, this...
Latent Dirichlet Allocation (LDA) has been widely applied to text mining. LDA is a probabilistic topic model which processes documents as the probability distribution of topics. One challenging issue in application select optimal number topics model. This paper presents selection method considers density each and computes most unstable structure through an iteration process. Evaluation results show that proposed can generate automatically with small iterations.
Skeleton-based action recognition has recently received considerable attention. Current approaches to skeleton-based are typically formulated as one-hot classification tasks and do not fully exploit the semantic relations between actions. For example, "make victory sign" "thumb up" two actions of hand gestures, whose major difference lies in movement hands. This information is agnostic from categorical encoding classes but could be unveiled description. Therefore, utilizing description...
Accurate hyperspectral remote sensing information is essential for feature identification and detection. Nevertheless, the imaging mechanism poses challenges in balancing trade-off between spatial spectral resolution. Hardware improvements are cost-intensive depend on strict environmental conditions extra equipment. Recent methods have attempted to directly reconstruct from widely available multispectral images. However, fixed mapping approaches used previous reconstruction models limit...
© 2019, Springer Nature Switzerland AG. Image classification is a difficult machine learning task, where Convolutional Neural Networks (CNNs) have been applied for over 20 years in order to solve the problem. In recent years, instead of traditional way only connecting current layer with its next layer, shortcut connections proposed connect forward layers apart from which has proved be able facilitate training process deep CNNs. However, there are various ways build connections, it hard...
Bayesian Network is the popular and important data mining model for representing uncertain knowledge. For large scale it often too costly to learn accurate structure. To resolve this problem, much work has been done on migrating structure learning algorithms MapReduce framework. In paper, we introduce a distributed hybrid algorithm by combining advantages of constraint-based score-and-search-based algorithms. By reusing intermediate results MapReduce, greatly simplified computing got good in...
Aiming at the problem of poor detection performance under environment imbalanced type distribution, an intrusion model genetic algorithm to optimize weighted extreme learning machine based on stratified cross-validation (SCV-GA-WELM) is proposed. In order solve data types in subsets, SCV used ensure that distribution all subsets consistent, thus avoiding over-fitting. The traditional fitness function cannot small sample classification well. By designing a and giving high weight data, can be...
Multi-sensor image matching based on salient edges has broad prospect in applications, but it is difficult to extract of real multi-sensor images with noises fast and accurately by using common algorithms. According the analysis features edges, a novel detection algorithm its rapid calculation are proposed possibility fuzzy C-means (PFCM) kernel clustering two-dimensional vectors composed values gray texture. PFCM can overcome shortcomings that (FCM) sensitive (PCM) tends find identical...
Deep metric learning turns to be attractive in zero-shot image retrieval and clustering (ZSRC) task which a good embedding/metric is requested such that the unseen classes can distinguished well. Most existing works deem this "good" embedding just discriminative one race devise powerful objectives or hard-sample mining strategies for deep metrics. However, article, we first emphasize generalization ability also core ingredient of it largely affects performance settings as matter fact. Then,...
The goal of few-shot learning is to learn a model that can recognize novel classes based on one or few training data. It challenging mainly due two aspects: (1) it lacks good feature representation classes; (2) labeled data could not accurately represent the true distribution and thus it's hard decision function for classification. In this work, we use sophisticated network architecture better focus second issue. A continual local replacement strategy proposed address deficiency problem....
Quantum theory has attracted people's attention since it was proposed. Due to its unique advantages in information storage and processing, quantum processing become the most popular research field. also provides us with new methods or concepts for manipulation processing. The basic problems of classical physics are basically trying be solved a situation isolation from surrounding environment reduce complexity analysis problem, but system inevitably produces decoherence establishes close...
In order to achieve better control performance for process systems (PCSs) a plenty of advanced methods have been proposed recently and most them all need the support data generated during operation. While not much attention had paid issue anomaly detection PCSs, which may be main factor that prevents practical applications these methods. This paper general method online identifying anomalies mixed among PCSs. Converting one-class classification (OCC) can remit absence in Exploiting thought...