A5100774764- Topic Modeling
- Advanced Text Analysis Techniques
- Natural Language Processing Techniques
- Advanced Graph Neural Networks
- Computational and Text Analysis Methods
- Bayesian Methods and Mixture Models
- Text and Document Classification Technologies
- Statistical Methods and Inference
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Complex Network Analysis Techniques
- Remote-Sensing Image Classification
- Advanced Image and Video Retrieval Techniques
- Graph Theory and Algorithms
- Statistical Methods and Bayesian Inference
- Matrix Theory and Algorithms
- Mineral Processing and Grinding
- Web Data Mining and Analysis
- Semantic Web and Ontologies
- Neural Networks and Applications
- Adsorption and biosorption for pollutant removal
- Image and Signal Denoising Methods
- Machine Learning and Data Classification
- Image Retrieval and Classification Techniques
- Pain Management and Treatment
Liaoning University of Traditional Chinese Medicine
2025
Jiangsu University
2019-2025
Affiliated Hospital of Jiangsu University
2022-2025
Foshan University
2025
Nanyang Technological University
2024
Shandong Normal University
2024
Jining University
2024
Xidian University
2018-2022
East China Normal University
2021-2022
Ministry of Natural Resources
2022
ive document summarization is a comprehensive task including understanding and summary generation, in which area Transformer-based models have achieved the state-of-the-art performance. Compared with Transformers, topic are better at learning explicit semantics, hence could be integrated into Transformers to further boost their To this end, we rearrange explore semantics learned by model, then propose assistant (TA) three modules. TA compatible various user-friendly since i) plug-and-play...
We develop a recurrent gamma belief network (rGBN) for radar automatic target recognition (RATR) based on high-resolution range profile (HRRP), which characterizes the temporal dependence across cells of HRRP. The proposed rGBN adopts hierarchy distributions to build its deep generative model. For scalable training and fast out-of-sample prediction, we propose hybrid stochastic-gradient Markov chain Monte Carlo (MCMC) variational inference model perform posterior inference. To utilize label...
CITE-seq provides a powerful method for simultaneously measuring RNA and protein expression at the single-cell level. The integrated analysis of in identical cells is crucial revealing cellular heterogeneity. However, high experimental costs associated with limit its widespread application. In this paper, we propose scTEL, deep learning framework based on Transformer encoder layers, to establish mapping from sequenced unobserved same cells. This computation-based approach significantly...
Abstract Biochar is a promising technology for carbon storage and greenhouse gas (GHG) reduction, but optimizing it challenging due to the complexity of natural systems. Machine learning (ML) language processing (NLP) offer solutions through enhanced data analysis pattern recognition, ushering in new era biochar research. Graphical
In the face of escalating crisis water pollution, biochar-based hydrogel composites (BCGs) have emerged as a promising material for treatment, owing to their distinctive performance and environmental friendliness. These combine high specific surface area porous structure biochar with three-dimensional network hydrogel, demonstrating superior adsorption capacities ease recyclability within aquatic systems. This paper provides first overview BCGs synthesis methods, particular emphasis on...
Predicting mortality rates is a crucial issue in life insurance pricing and demographic statistics. Traditional approaches, such as the Lee-Carter model its variants, predict trends of using factor models, which explain variations from perspective ages, gender, regions, other factors. Recently, deep learning techniques have achieved great success various tasks shown strong potential for time-series forecasting. In this paper, we propose modified Transformer architecture predicting major...
As machine learning algorithms are increasingly deployed for high-impact automated decision-making, the presence of bias (in datasets or tasks) gradually becomes one most critical challenges in applications. Such range from race face recognition to gender hiring systems, where and can be denoted as sensitive attributes. In recent years, much progress has been made ensuring fairness reducing standard settings. Among them, fair representations with respect attributes attracted increasing...
To learn a deep generative model of multimodal data, we propose Poisson gamma belief network (mPGBN) that tightly couple the data different modalities at multiple hidden layers. The mPGBN unsupervisedly extracts nonnegative latent representation using an upward-downward Gibbs sampler. It imposes sparse connections between layers, making it simple to visualize process and relationships features modalities. Our experimental results on bi-modal consisting images tags show can easily impute...
The rapid development of mobile Internet has brought new opportunities to college education and teaching. Taking university English teaching as the research object, this paper analyses characteristics in classroom, WeChat platform, learning app other forms teaching, changes influence on its application According actual situation, a comprehensive evaluation system based process results was established. show that model innovative examination can effectively improve efficiency students'...
Large Language Models (LLMs) have demonstrated impressive capability in many nature language tasks. However, the auto-regressive generation process makes LLMs prone to produce errors, hallucinations and inconsistent statements when performing multi-step reasoning. In this paper, we aim alleviate pathology by introducing Q*, a general, versatile agile framework for guiding decoding with deliberative planning. By learning plug-and-play Q-value model as heuristic function, our Q* can...
For text analysis, one often resorts to a lossy representation that either completely ignores word order or embeds each as low-dimensional dense feature vector. In this paper, we propose convolutional Poisson factor analysis (CPFA) directly operates on lossless processes the words in document sequence of high-dimensional one-hot vectors. To boost its performance, further gamma belief network (CPGBN) couples CPFA with via novel probabilistic pooling layer. forms into phrases and captures very...
Modulation recognition has always been an important task in the development of cognitive radio. At present, there are two main application methods for signal data, namely, directly using sequence and some conversions such as constellation diagram. In this paper, converted contour stella images adopted data source research. The deep learning method proposed, which is called Image-based CNN with Attention Model (ICAM). ICAM based on Residual Neural Network (ResNet). To evaluate performance...
Hierarchical topic models such as the gamma belief network (GBN) have delivered promising results in mining multi-layer document representations and discovering interpretable taxonomies. However, they often assume prior that topics at each layer are independently drawn from Dirichlet distribution, ignoring dependencies between both same across different layers. To relax this assumption, we propose sawtooth factorial embedding guided GBN, a deep generative model of documents captures semantic...
We propose a Bayesian generative model for incorporating prior domain knowledge into hierarchical topic modeling. Although embedded models (ETMs) and its variants have gained promising performance in text analysis, they mainly focus on mining word co-occurrence patterns, ignoring potentially easy-to-obtain hierarchies that could help enhance coherence. While several knowledge-based recently been proposed, are either only applicable to shallow or sensitive the quality of provided knowledge....
Abstract High-dimensional covariance matrix estimation plays a central role in multivariate statistical analysis. It is well-known that the sample singular when size smaller than dimension of variable, but estimate must be positive-definite. This motivates some modifications to preserve its efficient pairwise covariance. In this paper, we modify correlation using Bagging technique. The proposed estimator flexible for general continuous data. Under mild conditions, show theoretically can...
For document analysis, existing methods often resort to the representation that either discards word order information or projects each into a low-dimensional dense embedding vector. However, confined by data's sparsity and high-dimensionality, limited effort has been made explore semantic structures underlying formulates as sequence of one-hot vectors, especially in probabilistic modeling literature. To construct generative model for this type representation, we first develop convolutional...