A5100669567- Language, Discourse, Communication Strategies
- Multilingual Education and Policy
- Speech Recognition and Synthesis
- Discourse Analysis in Language Studies
- Speech and Audio Processing
- Traumatic Brain Injury and Neurovascular Disturbances
- Fault Detection and Control Systems
- Axon Guidance and Neuronal Signaling
- Neuroinflammation and Neurodegeneration Mechanisms
- Interpreting and Communication in Healthcare
- S100 Proteins and Annexins
- Intensive Care Unit Cognitive Disorders
- Anesthesia and Neurotoxicity Research
- Hate Speech and Cyberbullying Detection
- Neural dynamics and brain function
- Swearing, Euphemism, Multilingualism
- Tuberculosis Research and Epidemiology
- Natural product bioactivities and synthesis
- Cell Adhesion Molecules Research
- Pneumocystis jirovecii pneumonia detection and treatment
- Media Influence and Politics
- Speech and dialogue systems
- Topic Modeling
- Traumatic Brain Injury Research
- Memory and Neural Mechanisms
University of Science and Technology Beijing
2025
University of North Carolina at Chapel Hill
2022-2024
Lanzhou University of Technology
2023
Zhongda Hospital Southeast University
2023
Chinese Academy of Sciences
2022
Institute of High Energy Physics
2022
University of Chinese Academy of Sciences
2022
General Electric (United States)
2019
SUNY Downstate Health Sciences University
2018
East China Normal University
2018
Recently, end-to-end automatic speech recognition has become the mainstream approach in both industry and academia. To optimize system performance specific scenarios, Weighted Finite-State Transducer (WFST) is extensively used to integrate acoustic language models, leveraging its capacity implicitly fuse models within static graphs, thereby ensuring robust while also facilitating rapid error correction. However, WFST necessitates a frame-by-frame search of CTC posterior probabilities through...
Embedding-based retrieval models have made significant strides in retrieval-augmented generation (RAG) techniques for text and multimodal large language (LLMs) applications. However, when it comes to speech larage (SLLMs), these methods are limited a two-stage process, where automatic recognition (ASR) is combined with text-based retrieval. This sequential architecture suffers from high latency error propagation. To address limitations, we propose unified embedding framework that eliminates...
This paper investigates approximation-theoretic aspects of the in-context learning capability transformers in representing a family noisy linear dynamical systems. Our first theoretical result establishes an upper bound on approximation error multi-layer with respect to $L^2$-testing loss uniformly defined across tasks. demonstrates that logarithmic depth can achieve bounds comparable those least-squares estimator. In contrast, our second non-diminishing lower for class single-layer...
Abstract Background In the context of decreasing tuberculosis prevalence in China, we examined effectiveness screening household contacts patients. Methods A survey was conducted 2008. All 3,355 notified cases were with a questionnaire interview, chest X-ray and three sputum smear tests. The by comparing pulmonary or without presenting clinical symptoms against respective notification rates. Regression models used to evaluate factors associated tuberculosis. Results Of contacts, 92 members...
Background: Uncertainty continues to surround mild traumatic brain injury (mTBI) diagnosis, symptoms, prognosis, and outcome due in part a lack of objective biomarkers recovery. As mTBI gains recognition as serious public health epidemic, there is need identify risk factors, diagnostic tools, imaging help guide diagnosis management. Methods: One hundred eleven patients (15-50 years old) were enrolled acutely after followed with up four standardized serial assessments over 3 months. Each...
Analyzing a thread of online interaction, I apply conversation analysis and discursive psychology methods to explicate how experiences racism are reported contested by participants in interaction. The person reporting their experience (the reporter) applies commonsense knowledge assess the perpetrator's racist intent. Recipients report contest reporter's rights intent while managing lack independent access encounter. In milder contestations, they cast doubt avoiding assessing situation...
Animal contextual fear conditioning (CFC) models are the most-studied forms used to explore neural substances of posttraumatic stress disorder (PTSD). In addition well-recognized hippocampal–amygdalar system, retrosplenial cortex (RSC) is getting more and attention due substantial involvement in CFC, but with a poor understanding specific roles its two major constituents—dysgranular (RSCd) granular (RSCg). The current study sought identify their underlying brain network mechanisms during...
This study examines interviewers’ practices in managing conflicting institutional expectations of neutrality and rapport interview interaction which are complicated by the normative expectation antiracism. By applying conversation analysis membership categorization to sequences after interviewees’ possibly racist talk, we demonstrate that interviewer deploys various interactional resources orient antiracism, while maintaining a neutralistic stance sustaining with interviewees. The...
Foundation models for natural language processing, powered by the transformer architecture, exhibit remarkable in-context learning (ICL) capabilities, allowing pre-trained to adapt downstream tasks using few-shot prompts without updating their weights. Recently, transformer-based foundation have also emerged as versatile tools solving scientific problems, particularly in realm of partial differential equations (PDEs). However, theoretical foundations ICL capabilities these remain largely...
This article adopts an interactionist approach to examine the normative features and everyday reproduction of commonsense knowledge regarding race racism in social interaction among Asian members. Specifically, applying conversation analysis membership categorization forum interactions online American communities, I focus on interactional phenomenon where recipients a report, responding make relevant perpetrator(s)’ that is not identified report. identify three sets practices adopted by...
This paper examines members’ practices involved in accounting for discrimination against their own group interaction. Applying membership categorization analysis to interviews conducted with Chinese international students Japan, I show that categories and category-based knowledge constitute crucial resources target-of-discrimination making intelligible reasonable. Specifically, interviewee accounts by bounding certain activities/traits the category being discriminated against. Meanwhile,...