- Topic Modeling
- Advanced Graph Neural Networks
- Natural Language Processing Techniques
- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Text and Document Classification Technologies
- Speech Recognition and Synthesis
- Image Enhancement Techniques
- Smart Grid and Power Systems
- Web Data Mining and Analysis
- Complex Network Analysis Techniques
- High-Voltage Power Transmission Systems
- Underwater Vehicles and Communication Systems
- Music and Audio Processing
- Traffic control and management
- Brain Tumor Detection and Classification
- Transportation Planning and Optimization
- Technology and Security Systems
- Traffic Prediction and Management Techniques
- Video Surveillance and Tracking Methods
- Speech and Audio Processing
- Sentiment Analysis and Opinion Mining
- Advanced Text Analysis Techniques
- Rough Sets and Fuzzy Logic
- Advanced Computational Techniques and Applications
Xinjiang University
2024
Wuhan University of Technology
2012-2024
Shaoxing People's Hospital
2024
University of Science and Technology of China
2010-2022
Shijiazhuang Tiedao University
2022
Nanjing University of Aeronautics and Astronautics
2022
Guangzhou College of Commerce
2021
Stanford University
2020-2021
JDSU (United States)
2019-2021
Guangzhou Vocational College of Science and Technology
2021
Knowledge graph embedding has been an active research topic for knowledge base completion, with progressive improvement from the initial TransE, TransH, DistMult et al to current state-of-the-art ConvE. ConvE uses 2D convolution over embeddings and multiple layers of nonlinear features model graphs. The can be efficiently trained scalable large However, there is no structure enforcement in space recent convolutional network (GCN) provides another way learning node by successfully utilizing...
Document-level relation extraction (RE) poses new challenges compared to its sentence-level counterpart. One document commonly contains multiple entity pairs, and one pair occurs times in the associated with possible relations. In this paper, we propose two novel techniques, adaptive thresholding localized context pooling, solve multi-label multi-entity problems. The replaces global threshold for classification prior work a learnable entities-dependent threshold. pooling directly transfers...
Existing enhancement methods are empirically expected to help the high-level end computer vision task: however, that is observed not always be case in practice. We focus on object or face detection poor visibility enhancements caused by bad weathers (haze, rain) and low light conditions. To provide a more thorough examination fair comparison, we introduce three benchmark sets collected real-world hazy, rainy, low-light conditions, respectively, with annotated objects/faces. launched UG <sup...
Multi-hop reading comprehension (RC) across documents poses new challenge over single-document RC because it requires reasoning multiple to reach the final answer. In this paper, we propose a model tackle multi-hop problem. We introduce heterogeneous graph with different types of nodes and edges, which is named as Heterogeneous Document-Entity (HDE) graph. The advantage HDE that contains granularity levels information including candidates, entities in specific document contexts. Our proposed...
Interpretable multi-hop reading comprehension (RC) over multiple documents is a challenging problem because it demands reasoning information sources and explaining the answer prediction by providing supporting evidences. In this paper, we propose an effective interpretable Select, Answer Explain (SAE) system to solve multi-document RC problem. Our first filters out answer-unrelated thus reduce amount of distraction information. This achieved document classifier trained with novel pairwise...
Distance-based knowledge graph embeddings have shown substantial improvement on the link prediction task, from TransE to latest state-of-the-art RotatE. However, complex relations such as N-to-1, 1-to-N and N-to-N still remain challenging predict. In this work, we propose a novel distance-based approach for prediction. First, extend RotatE 2D domain high dimensional space with orthogonal transforms model relations. The transform embedding keeps capability modeling symmetric/anti-symmetric,...
Self-attention mechanism in graph neural networks (GNNs) led to state-of-the-art performance on many representation learning tasks. Currently, at every layer, attention is computed between connected pairs of nodes and depends solely the two nodes. However, such does not account for that are directly but provide important network context. Here we propose Multi-hop Attention Graph Neural Network (MAGNA), a principled way incorporate multi-hop context information into layer computation. MAGNA...
There is a lack of consensus as to the best way monitoring psoriasis severity in clinical trials. The Psoriasis Area and Severity Index (PASI) most frequently used system Physician's Global Assessment (PGA) also often used. However, both instruments have some drawbacks neither has been fully evaluated terms 'validity' 'reliability' rating scale. Lattice System (LS-PGA) scale recently developed address disadvantages PASI PGA.To evaluate inter-rater intrarater reliability PASI, PGA LS-PGA.On...
Article Free Access Share on Combining supervised learning with color correlograms for content-based image retrieval Authors: Jing Huang Department of Computer Science, Cornell University, Ithaca, NY NYView Profile , S. Ravi Kumar Mandar Mitra Authors Info & Claims MULTIMEDIA '97: Proceedings the fifth ACM international conference MultimediaNovember 1997 Pages 325–334https://doi.org/10.1145/266180.266383Published:01 November 1997Publication History 94citation876DownloadsMetricsTotal...
Multiple instance learning (MIL) aims to learn the mapping between a bag of instances and bag-level label. In this paper, we propose new end-to-end graph neural network (GNN) based algorithm for MIL: treat each as use GNN embedding, in order explore useful structural information among bags. The final representation is fed into classifier label prediction. Our first attempt MIL. We empirically show that proposed achieves state art performance on several popular MIL data sets without losing...
Temporal traffic prediction is critical for ITS yet remains challenging in handling complex spatio-temporal dynamics of systems. The continuous data (e.g., flow, and speed) from various channels nodes a network are coupled with each other over the time points channel, spatially between nodes, jointly both spatial temporal dimensions. Such multi-aspect couplings reflect conditions real-life system evolve movement dynamics. recent studies formulate by high-profile graph neural networks....
The rapid and effective identification of pathogens in patients with pulmonary infections has posed a persistent challenge medicine, conventional microbiological tests (CMTs) proving time-consuming less sensitive, hindering early diagnosis respiratory infections. While there been some research on the clinical performance targeted sequencing technologies, limited focus directed toward bronchoalveolar lavage fluid (BALF). This study primarily evaluates pathogen detection capabilities...
Commonsense knowledge graph (CKG) is a special type of (KG), where entities are composed free-form text. Existing CKG completion methods focus on transductive learning setting, all the present during training. Here, we propose first inductive setting for completion, unseen may appear at test time. We emphasize that crucial CKGs, because frequently introduced due to fact CKGs dynamic and highly sparse. InductivE as framework targeted task. ensures capability by directly computing entity...
Self-attention mechanism in graph neural networks (GNNs) led to state-of-the-art performance on many representation learning tasks. Currently, at every layer, attention is computed between connected pairs of nodes and depends solely the two nodes. However, such does not account for that are directly but provide important network context. Here we propose Multi-hop Attention Graph Neural Network (MAGNA), a principled way incorporate multi-hop context information into layer computation. MAGNA...
This paper presents a Dynamic Vision Sensor with few interesting features. We directly readout the voltage on logarithmic photo detector as intensity information of fired pixel. The computer is able to command sensor produce full-frame picture at whatever time it needs. was implemented using AMS 0.35 μm 2P4M Opto process an array 384 × 320 pixels. Each pixel occupies footprint 30 μm2, 12% fill factor.
Monocular depth estimation relied on RGB images is an important ill posed problem in ithe system of computer vision. Recently, people use the method deep learning to discuss this Most existing monocular algorithms convolution neural network. Depth based 2D has applications image segmentation, 3D object detection, robot navigation, tracking and autonomous driving. This paper gives a brief overview problem, reviews, evaluates discusses learning, looks forward direction further research face...
Multi-hop reading comprehension (RC) across documents poses new challenge over single-document RC because it requires reasoning multiple to reach the final answer. In this paper, we propose a model tackle multi-hop problem. We introduce heterogeneous graph with different types of nodes and edges, which is named as Heterogeneous Document-Entity (HDE) graph. The advantage HDE that contains granularity levels information including candidates, entities in specific document contexts. Our proposed...
Aspect-level sentiment classification aims to identify the polarity towards a specific aspect term in sentence. Most current approaches mainly consider semantic information by utilizing attention mechanisms capture interactions between context and term. In this paper, we propose employ graph convolutional networks (GCNs) on dependency tree learn syntax-aware representations of terms. GCNs often show best performance with two layers, deeper do not bring additional gain due over-smoothing...
Digital archives have emerged as the pre-eminent method for capturing human experience. Before such can be used efficiently, their contents must described. The NSF-funded MALACH project aims to provide improved access large spoken by advancing state-of-the-art in automated speech recognition (ASR), Information Retrieval (IR) and related technologies [1,2] multiple languages. This paper describes ASR research English corpus. corpus consists of unconstrained, natural filled with disfluencies,...
Recently, there has been significant progress in the development of large models. Following success ChatGPT, numerous language models have introduced, demonstrating remarkable performance. Similar advancements also observed image generation models, such as Google's Imagen model, OpenAI's DALL-E 2, and stable diffusion which exhibited impressive capabilities generating images. However, similar to these still encounter unresolved challenges. Fortunately, availability open-source their...