A5100754485- Topic Modeling
- Multimodal Machine Learning Applications
- Natural Language Processing Techniques
- Advanced Image and Video Retrieval Techniques
- Video Analysis and Summarization
- Web Data Mining and Analysis
- Text and Document Classification Technologies
- Mass Spectrometry Techniques and Applications
- Expert finding and Q&A systems
- Human Pose and Action Recognition
- Biomedical Text Mining and Ontologies
- Caching and Content Delivery
- Advanced Text Analysis Techniques
- Advanced Chemical Physics Studies
- Atmospheric chemistry and aerosols
- Spectroscopy and Laser Applications
- Sentiment Analysis and Opinion Mining
- Domain Adaptation and Few-Shot Learning
- Recommender Systems and Techniques
- Data Stream Mining Techniques
- Atmospheric Ozone and Climate
- Data Management and Algorithms
- Information Retrieval and Search Behavior
- Semantic Web and Ontologies
- Aquatic Ecosystems and Phytoplankton Dynamics
China University of Geosciences (Beijing)
2023-2025
Microsoft Research Asia (China)
2015-2024
Chinese PLA General Hospital
2015-2024
National University of Singapore
2024
Zhejiang Normal University
2024
Huaibei Normal University
2015-2024
Chinese Academy of Sciences
2006-2023
Shenyang Jianzhu University
2023
The Graduate Center, CUNY
2023
University of Chinese Academy of Sciences
2005-2022
In recent years, artificial intelligence (AI) has made incredible progress. Advanced foundation models such as ChatGPT can offer powerful conversation, in-context learning, and code generation abilities for a broad range of open-domain tasks. They also generate high-level solution outlines domain-specific tasks based on their acquired common-sense knowledge. Nonetheless, they still face difficulties in specialized because lack sufficient data during pretraining make errors neural network...
To build Video Question Answering (VideoQA) systems capable of assisting humans in daily activities, seeking answers from long-form videos with diverse and complex events is a must. Existing multi-modal VQA models achieve promising performance on images or short video clips, especially the recent success large-scale pre-training. However, when extending these methods to videos, new challenges arise. On one hand, using dense sampling strategy computationally prohibitive. other relying sparse...
Actual evapotranspiration modeling is providing useful information for researchers and resource managers in agriculture water resources around the world. The performance of models depends on accuracy forcing inputs model parameters. We developed an improved approach to parameterization Operational Simplified Surface Energy Balance (SSEBop) using Forcing Normalizing Operation (FANO). SSEBop has two key parameters that define boundary conditions. FANO algorithm computes wet-bulb condition a...
Image-text matching is a vital cross-modality task in artificial intelligence and has attracted increasing attention recent years. Existing works have shown that learning semantic concepts useful to enhance image representation can significantly improve the performance of both image-to-text text-to-image retrieval. However, existing models simply detect from given image, which are less likely deal with long-tail occlusion concepts. Frequently co-occurred same scene, e.g. bedroom bed, provide...
Recently, Visual Question Answering (VQA) has emerged as one of the most significant tasks in multimodal learning it requires understanding both visual and textual modalities. Existing methods mainly rely on extracting image question features to learn their joint feature embedding via fusion or attention mechanism. Some recent studies utilize external VQA-independent models detect candidate entities attributes images, which serve semantic knowledge complementary VQA task. However, these...
Understanding narrated instructional videos is important for both research and real-world web applications. Motivated by video dense captioning, we propose a model to generate procedure captions from which are sequence of step-wise clips with description. Previous works on captioning learn segments without considering transcripts. We argue that transcripts in can enhance representation providing fine-grained complimentary semantic textual information. In this paper, introduce framework (1)...
Recent research on Large Language Models (LLMs) has led to remarkable advancements in general NLP AI assistants. Some studies have further explored the use of LLMs for planning and invoking models or APIs address more multi-modal user queries. Despite this progress, complex visual-based tasks still remain challenging due diverse nature visual tasks. This diversity is reflected two aspects: 1) Reasoning paths. For many real-life applications, it hard accurately decompose a query simply by...
Understanding user query intent is a crucial task to Question-Answering area. With the development of online health services, communities generate huge amount valuable medical data, where intention can be mined. However, queries posted by common users have many domain concepts and colloquial expressions, which make understanding intents very difficult. In this paper, we try find predict from realistic text queries. A CNN-LSTM attention model proposed intents, an unsupervised clustering...
According to the website AcronymFinder.com which is one of world's largest and most comprehensive dictionaries acronyms, an average 37 new human-edited acronym definitions are added every day. There 379,918 acronyms with 4,766,899 on that site up now, each has 12.5 average. It a very important research topic identify what exactly means in given context for document comprehension as well retrieval. In this paper, we propose two word embedding based models disambiguation. Word represent words...
Knowlege is important for text-related applications. In this paper, we introduce Microsoft Concept Graph, a knowledge graph engine that provides concept tagging APIs to facilitate the understanding of human languages. Graph built upon Probase, universal probabilistic taxonomy consisting instances and concepts mined from Web. We start by introducing construction through iterative semantic extraction procedures, which extract 2.7 million 1.68 billion Web pages. then use conceptualization...
Abstract The subduction of the Paleo‐Pacific plate, and tectonic extension mineralization in upper has been extensively studied last decades. How plate interactions along margins contributed to processes, however, remains controversial. This study focuses on Queshan Metamorphic Core Complex (MCC) Songjiagou gold deposit northern Jiaodong Peninsula, southeastern North China Craton, uncover processes factors Early Cretaceous mineralization. It is shown that MCC experienced early nearly...
In this paper, we focus on the imbalance issue, which is rarely studied in aspect term extraction and sentiment classification when regarding them as sequence labeling tasks. Besides, previous works usually ignore interaction between terms polarities. We propose a GRadient hArmonized CascadEd model (GRACE) to solve these problems. Specifically, cascaded module developed enhance interchange improve attention of tokens The polarities designed depend generated labels. To alleviate extend...
Understanding user intent from her sequential search behaviors, i.e. predicting the of each query in a session, is crucial for modern Web engines. However, due to huge number behavior variables and coarse level labels defined by human editors, it very difficult directly model behavioral dynamics or sessions. In this paper, we propose novel Sparse Hidden-Dynamic Conditional Random Fields (SHDCRF) learning their Through incorporating proposed hidden state variables, SHDCRF aims learn...
Abstract Human embryonic stem (hES) cells are typically maintained on mouse fibroblast (MEF) feeders or with MEF‐conditioned medium. However, these xenosupport systems greatly limit the therapeutic applications of hES because risk cross‐transfer animal pathogens. The cell niche is a unique tissue microenvironment that regulates self‐renewal and differentiation cells. Recent evidence suggests localized in low oxygen. We hypothesized hypoxia could maintain undifferentiated phenotype have...
With the development of science and technology, consumers not only require food to be safe, but also them keep original natural flavor nutritional value as well, while traditional chemical storage method has been increasingly unable satisfy consumers’ demand. When compared with method, physical technology more obvious advantage. This article introduces some commonly-used methods modern methods. Key words : New plasma electrolyzed reduction water super ice-temperature
Question retrieval, which aims to find similar versions of a given question, is playing pivotal role in various question answering (QA) systems. This task quite challenging, mainly regard five aspects: synonymy, polysemy, word order, length, and data sparsity. In this article, we propose unified framework simultaneously handle these problems. We use the combined with corresponding concept information synonymy problem polysemous problem. Concept embedding are learned at same time from both...
In recent years, different commercial Weblog subscribing systems have been proposed to return stories from users. subscribed feeds. this paper, we propose a novel clustering-based RSS aggregator called as Clusgator System (RCS) for reading. Note that an feed may several topics. A user only be interested in subset of these addition there could many multiple feeds, which discuss similar topic perspectives. but do not know how collect all feeds related topic. contrast previous works, cluster...