A5074226898- Advanced Graph Neural Networks
- Complex Network Analysis Techniques
- Text and Document Classification Technologies
- Retinal Imaging and Analysis
- Glaucoma and retinal disorders
- Bioinformatics and Genomic Networks
- Retinal Diseases and Treatments
- Graph Theory and Algorithms
- Advanced Image and Video Retrieval Techniques
- Mental Health Research Topics
- Domain Adaptation and Few-Shot Learning
- Image Retrieval and Classification Techniques
- Complex Systems and Time Series Analysis
- Face and Expression Recognition
- COVID-19 epidemiological studies
- Generative Adversarial Networks and Image Synthesis
- Music and Audio Processing
- Advanced Clustering Algorithms Research
- Topic Modeling
- Spam and Phishing Detection
- Innovative Educational Techniques
- Remote-Sensing Image Classification
- Opinion Dynamics and Social Influence
- Statistical Methods and Inference
- Recommender Systems and Techniques
The University of Tokyo
2019-2024
Hong Kong Polytechnic University
2017-2023
Beijing University of Posts and Telecommunications
2015
Anhui Polytechnic University
2009-2011
Anhui Business College
2007
Network embedding is increasingly employed to assist network analysis as it effective learn latent features that encode linkage information. Various methods have been proposed, but they are only designed for a single scenario. In the era of big data, different types related information can be fused together form coupled heterogeneous network, which consists two sub-networks connected by inter-network edges. this scenario, edges act comple- mentary in presence intra-network ones. This...
Link Prediction has been an important task for social and information networks. Existing approaches usually assume the completeness of network structure. However, in many real-world networks, links node attributes can be partially observable. In this paper, we study problem Cross View (CVLP) on observable where focus is to recommend nodes with only (or vice versa). We aim bridge gap by learning a robust consensus link-based attribute-based representations so that become comparable latent...
Knowledge Graph (KG) errors introduce non-negligible noise, severely affecting KG-related downstream tasks. Detecting in KGs is challenging since the patterns of are unknown and diverse, while ground-truth labels rare or even unavailable. A traditional solution to construct logical rules verify triples, but it not generalizable different have distinct with domain knowledge involved. Recent studies focus on designing tailored detectors ranking triples based KG embedding loss. However, they...
Abstract Multi-label learning deals with data examples which are associated multiple class labels simultaneously. Despite the success of existing approaches to multi-label learning, there is still a problem neglected by researchers, i.e., not only some values observed missing, but also completely unobserved for training data. We refer as missing and , argue that it necessary discover these in order mine useful knowledge make deeper understanding what behind In this paper, we propose new...
There are increasing interests in learning low-dimensional and dense node representations from the network structure which is usually high-dimensional sparse. However, most existing methods fail to consider semantic meanings of links. Different links may have different because similarities between two nodes can be different, e.g., share common neighbors similar demonstrated node-generated content. In this paper, former type referred as structure-close while latter content-close These types...
Network embedding fills the gap of applying tuple-based data mining models to networked datasets through learning latent representations or embeddings. However, it may not be likely associate embeddings with physical meanings just as name, embedding, literally suggests. Hence, built on interpretable. In this paper, we thus propose learn identity and interest embeddings, where user includes demographic affiliation information, is demonstrated by activities topics users are interested in. With...
Prediction of glaucomatous visual field loss has significant clinical benefits because it can help with early detection glaucoma as well decision-making for treatments. Glaucomatous is conventionally captured through sensitivity (VF ) measurement, which costly and time-consuming. Thus, existing approaches mainly predict future VF utilizing limited data collected in the past. Recently, optical coherence tomography (OCT) been adopted to measure retinal layers thickness (RT considerably more...
As there are various data mining applications involving network analysis, embedding is frequently employed to learn latent representations or embeddings that encode the structure. However, existing models only designed for a single scenario. It common nodes can have multiple types of relationships in big era, which results networks, e.g., social networks and gene regulatory networks. Jointly thus may make network-specific more comprehensive complete as same node expose similar complementary...
<title>Abstract</title> Graph data augmentation (GDA), which manipulates graph structure and/or attributes, has been demonstrated as an effective method for improving the generalization of neural networks on semi-supervised node classification. As a technique, label-preservation is critical, that is, labels should not change after manipulation. However, most existing methods overlook requirements. Determining label-preserving nature GDA highly challenging, owing to non-Euclidean structure....
Abstract Graph data augmentation (GDA), which manipulates graph structure and/or attributes, has been demonstrated as an effective method for improving the generalization of neural networks on semi-supervised node classification. As a technique, label preservation is critical, that is, labels should not change after manipulation. However, most existing methods overlook requirements. Determining label-preserving nature GDA highly challenging, owing to non-Euclidean structure. In this study,...
Network embedding has been widely employed in networked data mining applications as it can learn low-dimensional and dense node representations from the high-dimensional sparse network structure. While most existing methods only model proximity between two nodes regardless of order proximity, this paper proposes to explicitly multi-node proximities which be observed practice, e.g., multiple researchers coauthor a paper, genes co-express protein. Explicitly modeling is important because some...
We constructed a multitask learning model (latent space linear regression and deep [LSLR-DL]) in which the 2 tasks of cross-sectional predictions (using OCT) visual field (VF; central 10°) longitudinal progression VF (30°) were performed jointly via sharing (DL) component such that information from both was used an auxiliary manner (The Association for Computing Machinery's Special Interest Group on Knowledge Discovery Data Mining [SIGKDD] 2021). The purpose current study to investigate...
Network embedding has been increasingly employed in networked data mining applications as it is effective to learn node embeddings that encode the network structure. Existing models usually a single for each node. In practice, person may interact with others different roles, such interacting schoolmates student, and colleagues an employee. Obviously, roles exhibit characteristics or features. Hence, only learning responsible all not appropriate. this paper, we thus introduce concept of...
Existing multi-label learning (MLL) approaches mainly assume all the labels are observed and construct classification models with a fixed set of target (known labels). However, in some real applications, multiple latent may exist outside this hide data, especially for large-scale data sets. Discovering exploring hidden not only find interesting knowledge but also help us to build more robust model. In paper, novel approach named DLCL (i.e., Latent Class Labels MLL) is proposed which can...
Spatial attention has been introduced to convolutional neural networks (CNNs) for improving both their performance and interpretability in visual tasks including image classification. The essence of the spatial is learn a weight map which represents relative importance activations within same layer or channel. All existing mechanisms are local attentions sense that maps image-specific. However, medical field, there cases all images should share because set record kind symptom related object...
Abstract We are concerned with the issue of detecting changes and their signs from a data stream. For example, when given time series COVID-19 cases in region, we may raise early warning signals an epidemic by data. propose novel methodology to address this issue. The key idea is employ new information-theoretic notion, which call differential minimum description length change statistics (D-MDL), for measuring scores sign. first give fundamental theory D-MDL. then demonstrate its...
We study the cold-start link prediction problem where edges between vertices is unavailable by learning vertex-based similarity metrics. Existing metric methods for fail to consider communities which can be observed in many real-world social networks. Because different usually exhibit intra-community homogeneities, a global not appropriate. In this paper, we thus propose learn community-specific metrics via joint community detection. Experiments on three networks show that homogeneities well...
Multi-label image classification (MLIC) is a fundamental and practical task, which aims to assign multiple possible labels an image. In recent years, many deep convolutional neural network (CNN) based approaches have been proposed model label correlations discover semantics of learn semantic representations images. This paper advances this research direction by improving both the modeling learning representations. On one hand, besides local each label, we propose further explore global...