A5101456529- Natural Language Processing Techniques
- Topic Modeling
- Semantic Web and Ontologies
- Advanced Text Analysis Techniques
- Neural Networks and Applications
- Speech and dialogue systems
- Text and Document Classification Technologies
- Rough Sets and Fuzzy Logic
- Advanced Image Fusion Techniques
- Advanced Image and Video Retrieval Techniques
- Image and Signal Denoising Methods
- Information Retrieval and Search Behavior
- Text Readability and Simplification
- Image Retrieval and Classification Techniques
- Advanced Image Processing Techniques
- Handwritten Text Recognition Techniques
- Remote-Sensing Image Classification
- Fuzzy Logic and Control Systems
- Speech Recognition and Synthesis
- Web Data Mining and Analysis
- Second Language Acquisition and Learning
- Mathematics, Computing, and Information Processing
- Data Mining Algorithms and Applications
- Advanced Computational Techniques and Applications
- Evolutionary Algorithms and Applications
Fudan University
2025
Shandong University
2011-2025
Harbin Institute of Technology
2020-2023
Zhejiang University of Technology
2012-2022
Ryukoku University
2010-2022
Peng Cheng Laboratory
2022
Beijing University of Posts and Telecommunications
2010-2021
Beijing Normal University
2005-2020
University of Science and Technology of China
2019-2020
Chinese Institute for Brain Research
2019-2020
Abstract Psychiatric disorders, including schizophrenia (SCZ), bipolar disorder (BD), and major depressive (MDD), share clinical neurobiological features. Because previous investigations of functional dysconnectivity have mainly focused on single the transdiagnostic alterations in connectome architecture brain remain poorly understood. We collected resting-state magnetic resonance imaging data from 512 participants, 121 with SCZ, 100 BD, 108 MDD, 183 healthy controls. Individual connectomes...
In this paper, we explore the spatiospectral image super-resolution (SSSR) task, <i>i.e.</i>, joint spatial and spectral super-resolution, which aims to generate a high resolution hyperspectral (HR-HSI) from low multispectral (LR-MSI). To tackle such severely ill-posed problem, one straightforward but inefficient way is sequentially perform single (SISR) network followed by (SSR) in two-stage manner or reverse order. propose model-based deep learning for SSSR named unfolding (US3RN), not...
With the rapid development of deep neural networks, cross-modal hashing has made great progress. However, information different types data is asymmetrical, that to say, if resolution an image high enough, it can reproduce almost 100% real-world scenes. text usually carries personal emotion and not objective so we generally think will be much richer than text. Although most existing methods unify semantic feature extraction hash function learning modules for end-to-end learning, they ignore...
Identifying reliable synthesis pathways in materials chemistry is a complex task, particularly polymer science, due to the intricate and often non-unique nomenclature of macromolecules. To address this challenge, we propose an agent system that integrates large language models (LLMs) knowledge graphs (KGs). By leveraging LLMs' powerful capabilities for extracting recognizing chemical substance names, storing extracted data structured graph, our fully automates retrieval relevant literatures,...
Abstract Identifying reliable synthesis pathways in materials chemistry is a complex task, particularly polymer science, due to the intricate and often nonunique nomenclature of macromolecules. To address this challenge, an agent system that integrates large language models (LLMs) knowledge graphs proposed. By leveraging LLMs' powerful capabilities for extracting recognizing chemical substance names, storing extracted data structured graph, fully automates retrieval relevant literature,...
High spatial resolution and high spectral images (HR-HSIs) are widely applied in geosciences, medical diagnosis, beyond. However, how to get with both is still a problem be solved. In this paper, we present deep spatial-spectral feature interaction network (SSFIN) for reconstructing an HR-HSI from low-resolution multispectral image (LR-MSI), e.g., RGB image. particular, introduce two auxiliary tasks, i.e., super-resolution (SR) SR help the recover better. Since higher can provide more...
One of the tasks Non-Intrusive Load Monitoring (NILM) is load identification, which aims to extract and classify altered electrical signals after switching events are detected. In this subtask, representative distinguishable signatures essential. At present, literature approach characterize appliances mainly based on manual feature engineering. However, performance obtained by way limited. paper, we propose a novel signature construction method utilizing deep learning techniques....
This paper describes named entity (NE) extraction based on a maximum entropy (M. E.) model and transformation rules. There are two types of entities when focusing the relationship between morphemes NEs as defined in NE task IREX competition held 1999. Each consists one or more morphemes, includes substring morpheme. We extract former type by using M. E. model. then latter applying rules to text.
Bipolar I disorder (BD-I) is associated with a high risk of suicide attempt; however, the neural circuit dysfunction that confers suicidal vulnerability in individuals this remains largely unknown. Resting-state functional magnetic resonance imaging (rs-fMRI) allows non-invasive mapping brain connectivity. The current study used an unbiased voxel-based graph theory analysis rs-fMRI to investigate intrinsic networks BD-I patients and without attempt.A total 30 attempt (attempter group), 82...
To solve the ill-posed problem of hyperspectral image super-resolution (HSISR), a usually method is to use prior information images (HSIs) as regularization term constrain objective function. Modelbased methods using hand-crafted priors cannot fully characterize properties HSIs. Learning-based convolutional neural network (CNN) learn implicit However, learning ability CNN limited, it only considers spatial characteristics HSIs and ignores spectral characteristics, convolution not effective...
Image fusion is famous as an alternative solution to generate one high-quality image from multiple images in addition restoration a single degraded image. The essence of integrate complementary information source images. Existing methods struggle with generalization across various tasks and often require labor-intensive designs, which it difficult identify extract useful due the diverse requirements each task. Additionally, these develop highly specialized features for different downstream...
Robertson's 2-poisson information retrieve model does not use location and category information. We constructed a framework using in model. submitted two systems based on this to the IREX contest, Japanese language retrieval contest held Japan 1999. For precision A-judgement measure they scored 0.4926 0.4827, highest values among 15 teams 22 that participated contest. describe our comparative experiments done when various parameters were changed. These confirmed effectiveness of
The elastic-input neuro-tagger and hybrid tagger, combined with a neural network Brill's error-driven learning, have already been proposed to construct practical tagger using as little training data possible. When small Thai corpus is used for training, these taggers tagging accuracies of, respectively, 94.4% 95.5% (accounting only the ambiguous words that relate parts of speech). In this study, in order more accurate taggers, we developed new methods three different machine-learning...
This paper presents a part of speech (POS) neuro tagger which consists 3-layer perceptron with elastic input. Computer experiments show that the has an accuracy 94.4% for tagging ambiguous words when small Thai corpus 22,311 is used training. A series comparative further definitely far superior to statistical models including frequency model (a base-line model), local n-gram model, and HMM.
CALL evaluation is important because it the most efficient means to prove effectiveness. While both learning process and outcome should be investigated in empirical evaluation, precise relationship between two needs examined closely. Only by doing so can we identify useful design features that facilitate relevant user-computer interaction which lead an improved outcome. This study how certain user actions affect or predict receptive/productive vocabulary retention a computer-assisted (CAVL)...
Abstract In this paper, we propose a method to estimate the order of paragraphs by supervised machine learning. We use support vector (SVM) for The estimation paragraph is useful sentence generation and correction. proposed obtained high accuracy (0.84) in experiments first two an article. addition, it higher than baseline using performed feature analysis found that adnominals, conjunctions, dates were effective paragraphs, ratio new words similarity between preceding estimated all pairs paragraphs.
In this paper we describe a method of acquiring word order from corpora. Word is defined as the modifiers, or phrasal units called 'bunsetsu' which depend on same modifiee. The uses model automatically discovers what tendency in Japanese by using various kinds information and around target bunsetsus. This shows us to extent each piece contributes deciding tends be selected when several conflict. contribution rate efficiently learned within maximum entropy framework. performance trained can...