A5100782897- Topic Modeling
- Advanced Graph Neural Networks
- Natural Language Processing Techniques
- Recommender Systems and Techniques
- Speech Recognition and Synthesis
- Data Quality and Management
- Complex Network Analysis Techniques
- Speech and Audio Processing
- Music and Audio Processing
- Software Engineering Research
- Speech and dialogue systems
- Algorithms and Data Compression
- Micro and Nano Robotics
- Molecular Communication and Nanonetworks
- Domain Adaptation and Few-Shot Learning
- Spam and Phishing Detection
- Atomic and Molecular Physics
- Advanced Computing and Algorithms
- Constructed Wetlands for Wastewater Treatment
- Quantum Chromodynamics and Particle Interactions
- Text Readability and Simplification
- Brain Tumor Detection and Classification
- Spatial Cognition and Navigation
- Education and Learning Interventions
- Innovative Educational Techniques
Hubei University
2022-2025
Central China Normal University
2016-2022
Beijing University of Civil Engineering and Architecture
2017
Peking University Third Hospital
2013
Peking University
2013
Google (United States)
2011
Johns Hopkins University
2008-2010
Institute of Modern Physics
2010
Chinese Academy of Sciences
2010
Knowledge graph (KG) embedding aims to study the representation retain inherent structure of KGs. Graph neural networks (GNNs), as an effective technique, have shown impressive performance in learning embedding. However, KGs intrinsic property heterogeneity, which contains various types entities and relations. How address complex data aggregate multiple semantic information simultaneously is a critical issue. In this article, novel heterogeneous GNNs framework based on attention mechanism...
Knowledge graphs are large graph-structured knowledge bases with incomplete or partial information. Numerous studies have focused on graph embedding to identify the embedded representation of entities and relations, thereby predicting missing relations between entities. Previous models primarily regard (subject entity, relation, object entity) triplet as translational distance semantic matching in vector space. However, these only learn a few expressive features hard handle complex i.e.,...
Recent advances in text-based large language models (LLMs), particularly the GPT series and o1 model, have demonstrated effectiveness of scaling both training-time inference-time compute. However, current state-of-the-art TTS systems leveraging LLMs are often multi-stage, requiring separate (e.g., diffusion after LLM), complicating decision whether to scale a particular model during training or testing. This work makes following contributions: First, we explore train-time compute for speech...
Here we describe the utilization of flagellated bacteria as actuators to propel spherical liposomes by attaching liposome surface. Bacteria were stably attached using a cross-linking antibody. The effect number on propulsion speed was experimentally determined. effects bacterial bacteria–antibody–liposome complex stochastic. We demonstrated that liposomal mobility increased when attached, and correlated with bacteria.
Analysis of the genesis and evolution mechanism High Steep Dangerous Rocks (HSDRs) are crucial for enhancing capability geological disaster risk prevention control in reservoir areas. This study focuses on Longmen Rock Zone (LDRZ) Three Gorges Reservoir Area (TGRA), China, employs an integrated investigation system proposed. The LDRZ can be attributed to early influence Indosinian movement, Yanshan Himalayan movement. development joint fissures Longmenxia Anticline led presence stacked...
We describe a scalable decoder for parsing-based machine translation. e is written in Java and implements all the essential algorithms described (Chiang, 2007) (Li Khudanpur, 2008b): chart-parsing, n-gram language model integration, beamand cube-pruning, k-best extraction. Additionally, parallel distributed computing techniques are exploited to make it scalable. demonstrate experimentally that our more than 30 times faster baseline Python.
Hypergraphs are used in several syntax-inspired methods of machine translation to compactly encode exponentially many hypotheses. The hypotheses closest given reference translations therefore cannot be found via brute force, particularly for popular measures closeness such as BLEU. We develop a dynamic program extracting the so called oracle-best hypothesis from hypergraph by viewing it problem finding most likely under an n-gram language model trained only translations. further identify and...
We describe Joshua (Li et al., 2009a), an open source toolkit for statistical machine translation. implements all of the algorithms required translation via synchronous context free grammars (SCFGs): chart-parsing, n-gram language model integration, beam- and cube-pruning, k-best extraction. The also suffix-array grammar extraction minimum error rate training. It uses parallel distributed computing techniques scalability. provide a demonstration outline illustrating toolkit's features to...
Flagellated bacteria have been utilized as potential swimming micro-robotic bodies for propulsion of spherical liposome by attaching several on their surface. Liposome a drug delivery vehicle can contain biologically active compounds. In this work, the antibody binding technique is developed to attach liposome's Consequently, stochastic effect bacterial investigated analytically and experimentally. It shown that mobility with was higher than without bacteria. Experimental data matches well...
Recently, text-guided content generation has received extensive attention. In this work, we explore the possibility of text description-based speaker generation, i.e., using prompts to control process. Specifically, propose PromptSpeaker, a system. PromptSpeaker consists prompt encoder, zero-shot VITS, and Glow model, where encoder predicts prior distribution based on description samples from obtain semantic representation. The model subsequently converts representation into representation,...
Recently, the interactive learning environment has received more and attention. In this paper, we introduce a cloud-terminal integration platform, which is developed by our group, design blended mode, enables platform to be used learning. Finally, verify effect of using course "Digital teaching system development".