A5100386207- Topic Modeling
- Natural Language Processing Techniques
- Advanced Graph Neural Networks
- Data Quality and Management
- Sentiment Analysis and Opinion Mining
- Speech Recognition and Synthesis
- Domain Adaptation and Few-Shot Learning
- Text and Document Classification Technologies
- Machine Learning and ELM
- Emotion and Mood Recognition
- Biomedical Text Mining and Ontologies
- Computational Physics and Python Applications
- Text Readability and Simplification
- Multimodal Machine Learning Applications
- Bioinformatics and Genomic Networks
- SARS-CoV-2 and COVID-19 Research
- Machine Learning in Healthcare
- Machine Learning and Data Classification
- Machine Learning and Algorithms
- Cholinesterase and Neurodegenerative Diseases
- Educational Technology and Assessment
- Second Language Acquisition and Learning
- Speech and Audio Processing
- Software Engineering Research
- Music and Audio Processing
Beijing Jiaotong University
2013-2024
Nanjing Tech University
2024
Nankai University
2018-2023
Data Assurance and Communication Security
2021-2022
University of Florida
2022
Université de Montréal
2016
Département d'Informatique
2016
Dalian University of Technology
2012
Jiangxi Science and Technology Normal University
2008
Convolutional Neural Networks (CNNs) are effective models for reducing spectral variations and modeling correlations in acoustic features automatic speech recognition (ASR).Hybrid systems incorporating CNNs with Hidden Markov Models/Gaussian Mixture Models (HMMs/GMMs) have achieved the state-of-the-art various benchmarks.Meanwhile, Connectionist Temporal Classification (CTC) Recurrent (RNNs), which is proposed labeling unsegmented sequences, makes it feasible to train an 'end-to-end' system...
The Teacher Forcing algorithm trains recurrent networks by supplying observed sequence values as inputs during training and using the network's own one-step-ahead predictions to do multi-step sampling. We introduce Professor algorithm, which uses adversarial domain adaptation encourage dynamics of network be same when sampling from over multiple time steps. apply language modeling, vocal synthesis on raw waveforms, handwriting generation, image generation. Empirically we find that acts a...
Large Language Models (LLMs), like ChatGPT, have demonstrated vast potential but also introduce challenges related to content constraints and misuse. Our study investigates three key research questions: (1) the number of different prompt types that can jailbreak LLMs, (2) effectiveness prompts in circumventing LLM constraints, (3) resilience ChatGPT against these prompts. Initially, we develop a classification model analyze distribution existing prompts, identifying ten distinct patterns...
The success of emotional conversation systems depends on sufficient perception and appropriate expression emotions. In a real-world conversation, we firstly instinctively perceive emotions from multi-source information, including the emotion flow dialogue history, facial expressions, personalities speakers, then express suitable according to our personalities, but these multiple types information are insufficiently exploited in fields. To address this issue, propose heterogeneous graph-based...
ABSTRACT Objective To develop a large pretrained clinical language model from scratch using transformer architecture; systematically examine how models of different sizes could help 5 natural processing (NLP) tasks at linguistic levels. Methods We created corpus with >90 billion words narratives (>82 words), scientific literature (6 and general English text (2.5 words). developed GatorTron the BERT architecture including 345 million, 3.9 billion, 8.9 parameters, compared three existing...
Entity Alignment (EA) aims to find the equivalent entities between two Knowledge Graphs (KGs). Existing methods usually encode triples of as embeddings and learn align embeddings, which prevents direct interaction original information cross-KG entities. Moreover, they relational attribute an entity in heterogeneous embedding spaces, them from helping each other. In this paper, we transform both into unified textual sequences, model EA task a bi-directional entailment sequences Specifically,...
Baohang Zhou, Xiangrui Cai, Ying Zhang, Xiaojie Yuan. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.
Deep neural networks (DNNs) have been widely used in many fields due to their powerful representation learning capabilities. However, they are exposed serious threats caused by the increasing security issues. Adversarial examples were early discovered computer vision (CV) field when models fooled perturbing original inputs, and also exist natural language processing (NLP) community. unlike image, text is discrete semantic nature, making generation of adversarial attacks even more difficult....
Biomedical entity linking is an essential task in biomedical text processing, which aims to map mentions text, such as clinical notes, standard terms a given knowledge base. However, this challenging due the rarity of many entities real-world scenarios, often leads lack annotated data for them. Limited by understanding these unseen entities, traditional models suffer from multiple types errors. In paper, we propose novel latent feature generation framework BioFEG address challenges....
Multimodal named entity recognition (MNER) on social media is a challenging task which aims to extract entities in free text and incorporate images classify them into user-defined types. However, the annotation for demands mount of human efforts. The existing semi-supervised methods focus modal are utilized reduce labeling costs traditional NER. previous not efficient MNER. Because MNER defined combine information with image one needs consider mismatch between posted image. To fuse features...
We present a novel architecture, In-Database Entity Linking (IDEL), in which we integrate the analytics-optimized RDBMS MonetDB with neural text mining abilities. Our system design abstracts core tasks of most entity linking systems for MonetDB. To best our knowledge, this is first defacto implemented integrating entity-linking database. leverage ability to support in-database-analytics user defined functions (UDFs) Python. These call machine learning libraries mining, such as TensorFlow....
Machine reading comprehension with multi-hop reasoning always suffers from path breaking due to the lack of world knowledge, which results in wrong answer detection. In this paper, we analyze what knowledge previous work lacks, e.g., dependency relations and commonsense. Based on our analysis, propose a Multi-dimensional Knowledge enhanced Graph Network, named MKGN, exploits specific repair gap process. Specifically, approach incorporates not only entities through various graph neural...
Mongolian named entity recognition (NER) is not only one of the most crucial and fundamental tasks in natural language processing, but also an important step to improve performance downstream such as information retrieval, machine translation, dialog system. However, traditional NER models heavily rely on feature engineering. Even worse, complex morphological structure words makes data sparser. To alleviate engineering sparsity recognition, we propose a novel framework with Multi-Knowledge...
With the expanding application of Large Language Models (LLMs) in various domains, it becomes imperative to comprehensively investigate their unforeseen behaviors and consequent outcomes. In this study, we introduce systematically explore phenomenon "glitch tokens", which are anomalous tokens produced by established tokenizers could potentially compromise models' quality response. Specifically, experiment on seven top popular LLMs utilizing three distinct involving a totally 182,517 tokens....
A large number of studies have emerged for Multimodal Knowledge Graph Completion (MKGC) to predict the missing links in MKGs. However, fewer been proposed study inductive MKGC (IMKGC) involving emerging entities unseen during training. Existing approaches focus on learning textual entity representations, which neglect rich semantic information visual modality. Moreover, they aggregating structural neighbors from existing KGs, are usually limited. decoupled topology linkage and imply true...
Parameter-efficient transfer learning (PETL) has emerged as a flourishing research field for adapting large pre-trained models to downstream tasks, greatly reducing trainable parameters while grappling with memory challenges during fine-tuning. To address it, memory-efficient series (METL) avoid backpropagating gradients through the backbone. However, they compromise by exclusively relying on frozen intermediate outputs and limiting exhaustive exploration of prior knowledge from models....
Unlike ordinary port cities, the historical particularity, cultural diversity, regional inclusiveness, and unique natural human resources of Quanzhou in China have made its architecture Chinese architecture. It is a typical paradigm long-term coexistence mutual influence traditional overseas culture. The research on characteristics origin development helpful to sustainable local protection. This paper explores, organizes, analyzes architectural language Quanzhou.