A5100462722- Topic Modeling
- Advanced Graph Neural Networks
- Natural Language Processing Techniques
- Graph Theory and Algorithms
- Web Data Mining and Analysis
- Distributed and Parallel Computing Systems
- Domain Adaptation and Few-Shot Learning
- Stochastic Gradient Optimization Techniques
- Power Systems and Technologies
- Robotics and Sensor-Based Localization
- Biomedical Text Mining and Ontologies
- Speech and Audio Processing
- Music and Audio Processing
- Multimodal Machine Learning Applications
- Data Mining Algorithms and Applications
- Advanced Vision and Imaging
- Cerebrovascular and Carotid Artery Diseases
- Imbalanced Data Classification Techniques
- Human Pose and Action Recognition
- Higher Education and Teaching Methods
- Mobile Agent-Based Network Management
- Complex Network Analysis Techniques
- AI and Multimedia in Education
- Data Quality and Management
- Software System Performance and Reliability
Shanghai University of Electric Power
2022-2025
Bangor University
2025
Chinese Academy of Medical Sciences & Peking Union Medical College
2024
Chinese University of Hong Kong
2020-2024
Nanjing University of Science and Technology
2024
Southwest Forestry University
2024
University of Minnesota
2023
Beijing University of Posts and Telecommunications
2022-2023
Xi'an Polytechnic University
2023
Henan University
2023
Despite recent success in using the invariance principle for out-of-distribution (OOD) generalization on Euclidean data (e.g., images), studies graph are still limited. Different from images, complex nature of graphs poses unique challenges to adopting principle. In particular, distribution shifts can appear a variety forms such as attributes and structures, making it difficult identify invariance. Moreover, domain or environment partitions, which often required by OOD methods data, could be...
The graph Laplacian regularization term is usually used in semi-supervised representation learning to provide structure information for a model f(X). However, with the recent popularity of neural networks (GNNs), directly encoding A into model, i.e., f(A, X), has become more common approach. While we show that brings little-to-no benefit existing GNNs, and propose simple but non-trivial variant regularization, called Propagation-regularization (P-reg), boost performance GNN models. We formal...
Recently Graph Injection Attack (GIA) emerges as a practical attack scenario on Neural Networks (GNNs), where the adversary can merely inject few malicious nodes instead of modifying existing or edges, i.e., Modification (GMA). Although GIA has achieved promising results, little is known about why it successful and whether there any pitfall behind success. To understand power GIA, we compare with GMA find that be provably more harmful than due to its relatively high flexibility. However,...
The previously proposed cascaded inverse fast Fourier transform/fast transform (IFFT/FFT)-based point-to-multipoint (P2MP) flexible optical transceivers have the potential to equip future intensity modulation and direct detection (IMDD) access networks with excellent flexibility, adaptability, scalability upgradability. However, due their IFFT-based multi-channel aggregations, P2MP suffer high peak-to-average power ratios (PAPRs). To address technical challenge, this paper proposes a novel...
The high cost of communicating gradients is a major bottleneck for federated learning, as the bandwidth participating user devices limited. Existing gradient compression algorithms are mainly designed data centers with high-speed network and achieve $O(\sqrt{d} \log d)$ per-iteration communication at best, where $d$ size model. We propose hyper-sphere quantization (HSQ), general framework that can be configured to continuum trade-offs between efficiency accuracy. In particular, ratio end,...
Graph neural networks (GNNs) have received much attention recently because of their excellent performance on graph-based tasks. However, existing research GNNs focuses designing more effective models without considering about the quality input data. In this paper, we propose self-enhanced GNN (SEG), which improves data using outputs for better semi-supervised node classification. As graph consist both topology and labels, improve from perspectives. For topology, observe that higher...
Background: Bevacizumab, a monoclonal antibody against vascular endothelial growth factor ligand, has shown survival benefits in the treatment of many types malignant tumors, including non-small-cell lung cancer (NSCLC). We conducted this systematic review and meta-analysis to investigate risk most clinically relevant adverse events related bevacizumab advanced NSCLC. Methods: Databases from PubMed, Web Science, Cochrane Library up August 2015, were searched identify studies. included...
Medical physicists are essential members of the radiation oncology team. Given increasing complexity radiotherapy delivery, it is important to ensure adequate training and staffing. The aim present study was update a similar survey from 2008 assess situation medical in large diverse Asia Pacific region.Between March July 2011, on profession practice (ROMPs) region performed. sent senior 22 countries. Replies were received countries that collectively represent more than half world's...
The urban heat island (UHI) effect caused by urbanization negatively impacts the ecological environment and human health. It is crucial for planning social development to monitor study its mechanism. Due spatial temporal resolution limitations, existing land surface temperature (LST) data obtained from remote sensing challenging meet long-term fine-scale mapping requirement. Given above situation, this paper introduced ResNet-based downscaling method make up deficiency applied it of thermal...
Abstract This paper presents a numerical study of the effects particle's shape and its interaction with surrounding fluid on mechanism sandpiles formation in air water, respectively. is motivated by fact that seabed sediments are predominantly deposited water consist non‐spherical particles. In our study, non‐linear contact model employed Discrete Element Method. At same time, void fraction drag force particles sharp corners further improved. Based above two advancements, an extended...
This paper presents the details of our system designed for Task 1 Multimodal Information Based Speech Processing (MISP) Challenge 2021. The purpose is to leverage both audio and video information improve environmental robustness far-field wake word spotting. In proposed system, firstly, we take advantage speech enhancement algorithms such as beamforming weighted prediction error (WPE) address multi-microphone conversational audio. Secondly, several data augmentation techniques are applied...
Large language models (LLM) have demonstrated remarkable capabilities in various biomedical natural processing (NLP) tasks, leveraging the demonstration within input context to adapt new tasks. However, LLM is sensitive selection of demonstrations. To address hallucination issue inherent LLM, retrieval-augmented (RAL) offers a solution by retrieving pertinent information from an established database. Nonetheless, existing research work lacks rigorous evaluation impact large on different NLP...
To investigate the demonstration in Large Language Models (LLMs) for clinical relation extraction. We focus on examining two types of adaptive demonstration: instruction prompting, and example prompting to understand their impacts effectiveness.
Recently, there has been a growing surge of interest in enabling machine learning systems to generalize well Out-of-Distribution (OOD) data. Most efforts are devoted advancing optimization objectives that regularize models capture the underlying invariance; however, often compromises process these OOD objectives: i) Many have be relaxed as penalty terms Empirical Risk Minimization (ERM) for ease optimization, while forms can weaken robustness original objective; ii) The also require careful...
Edit-distance-based string similarity search has many applications such as spell correction, data de-duplication, and sequence alignment. However, computing edit distance is known to have high complexity, which makes challenging for large datasets. In this paper, we propose a deep learning pipeline (called CNN-ED) that embeds into Euclidean fast approximate search. A convolutional neural network (CNN) used generate fixed-length vector embeddings dataset of strings the loss function...
Through the investigation and analysis of status influence factors post-earthquake disaster on urban transport system, this paper aims to introduce authors' study route choice for transporting exigency based principles timely, security economy under situation. To describes above three decision-making parameters in accordance with mathematics theory, it is necessary establish their objective functions through processing dimensionless technique weight aggregation so as set up a mathematical...
Graph neural networks (GNNs) have received much attention recently because of their excellent performance on graph-based tasks. However, existing research GNNs focuses designing more effective models without considering about the quality input data. In this paper, we propose self-enhanced GNN (SEG), which improves data using outputs for better semi-supervised node classification. As graph consist both topology and labels, improve from perspectives. For topology, observe that higher...
Graph contrastive learning algorithms have demonstrated remarkable success in various applications such as node classification, link prediction, and graph clustering. However, unsupervised learning, some pairs may contradict the truths downstream tasks thus decrease of losses on these undesirably harms performance tasks. To assess discrepancy between prediction ground-truth for pairs, we adapt expected calibration error (ECE) to learning. The analysis ECE motivates us propose a novel...