A5103050255- Speech Recognition and Synthesis
- Music and Audio Processing
- Speech and Audio Processing
- Natural Language Processing Techniques
- Banking stability, regulation, efficiency
- Global Financial Crisis and Policies
- Topic Modeling
- Advanced Computational Techniques and Applications
- Monetary Policy and Economic Impact
- Corporate Finance and Governance
- Credit Risk and Financial Regulations
- Economic theories and models
- Semantic Web and Ontologies
- Financial Reporting and Valuation Research
- Text and Document Classification Technologies
- Insurance and Financial Risk Management
- Italy: Economic History and Contemporary Issues
- Economic Theory and Policy
- Working Capital and Financial Performance
- Language, Metaphor, and Cognition
- Adrenal and Paraganglionic Tumors
- Cryptographic Implementations and Security
- Biomedical Text Mining and Ontologies
- Environmental and Agricultural Sciences
- Machine Fault Diagnosis Techniques
National University of Singapore
2008-2025
Xiamen University
2009-2025
Tianjin Normal University
2024
Minjiang University
2024
Xiamen University of Technology
2019-2023
Southwestern University of Finance and Economics
2010-2023
ORCID
2022
Wuhan University of Technology
2017-2021
University of Maryland, College Park
2019
Naval Aeronautical and Astronautical University
2018
Math word problem (MWP) is a challenging and critical task in natural language processing. Many recent studies formalize MWP as generation have adopted sequence-to-sequence models to transform descriptions mathematical expressions. However, expressions are prone minor mistakes while the objective does not explicitly handle such mistakes. To address this limitation, we devise new ranking for propose Generate & Rank, multi-task framework based on generative pre-trained model. By joint training...
Online condition monitoring and diagnosis systems play an important role in the modern manufacturing industry. This paper presents a novel method to diagnose degradation processes of multiple failure modes using modified hidden Markov model (MHMM) with variable state space. The proposed MHMM is combined statistical process control quickly detect occurrence unknown fault. allows space be adjusted updated identification new states. Hence, online assessment adaptive fault can simultaneously...
The tropics possess a higher level of biodiversity compared to other climatic regions on the Earth. have diverse ecosystems such as forests, savannas, mountains, wetlands, and coral reefs, which provide habitat for thousands species fauna flora. Tropical forests are one important tropical with high biodiversity. Humans modify forest landscapes large scale use them different purposes, including logging, mining, agriculture, pastures, urbanization, road construction. attain resources profits...
Abstract Defect detection of printed circuit board (PCB) is significant practical importance to ensure quality control in the production process. However, traditional defect methods suffer from limitations such as low accuracy and poor generalization ability. To tackle these issues, we propose a novel deep learning-based method for bare PCBs through multi-attention adaptive feature-enhancement fusion (AFF). First, utilize ResNext101 backbone feature extractor embed normalization-based...
Background The incidence of adrenal incidentalomas is increasing annually, and most types masses require surgical intervention. Accurate classification common based on tumor computed tomography (CT) images by radiologists or clinicians requires extensive experience often challenging, which increases the workload leads to unnecessary surgeries. There an urgent need for a fully automated, noninvasive, precise approach identification accurate masses. Objective This study aims enhance diagnostic...
This paper presents a new methodology for design and implementation of signal processing systems on system-on-chip (SoC) platforms. The is centered the use lightweight application programming interfaces applying principles dataflow at different layers abstraction. development processes integrated in our approach are software implementation, hardware hardware-software co-design, optimized mapping. proposed facilitates integration modules that involve heterogeneous languages As demonstration...
The paper proposes to determine sentence similarities from different aspects. Based on the information people get a sentence, Objects-Specified Similarity, Objects-Property Objects-Behavior Similarity and Overall are defined four Experiments show that proposed method makes similarity comparison more exactly give out reasonable result, which is similar people's comprehension meanings of sentences.
Based on respiratory sound production mechanisms, we study the relationship between airflow characteristics in bronchi and pressure spectrum curves to implement an end-to-end classification system with a feature-band attention module. First, analyse fluid-solid coupling simulations of execute acoustic obtain at level. Then, based bronchi, propose strategy refine features adaptive weights. In addition, introduce module ResNet-based networks squeeze-and-excitation block. Finally, perform...
The chirp signal, which is of large time bandwidth product, widely used in the current radar systems. However, high dynamic biases are brought into ranging when signal hypersonic target tracking, due to distance-Doppler coupling. In view situation, this letter proposes an unbiased tracking algorithm can eliminate influence at same tracking.
The central counterparties dominating the market for clearing of over-the-counter interest rate and credit derivatives are globally systemic. Employing methodologies similar to calculation banks' capital requirements against trading book exposures, this paper assesses sensitivity counterparties' required risk buffers, or requirements, a range model inputs. We find them be highly sensitive whether key parameters calibrated on point-in-time versus stress-period basis, tolerance metric...
An effective degradation indicator created from the general features is still a hotspot for condition monitoring of bearing. Since single feature–based cannot fully describe performance, some new indicators based on characteristics multiple are proposed using pattern recognition and dimensional reduction algorithms. Although most these can show incipient fault good tendency only across failure stage, also with oscillation, seldom them give whole health entire useful life. Hence, further...
Supervised deep learning has achieved tremendous success in speaker verification. However, models tend to overfit noisy labels when they are present the datasets. To mitigate detrimental effects of labels, this paper, we propose a novel <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Label Ensembling and Sample Selection</i> framework. Firstly, select with high confidence rankings as clean samples. Additionally, use predictions from different...
Linux Unified Key Setup (LUKS) is a popular disk encryption specification for operating system. It the first time to implement operations of LUKS with regard FPGA architectures, which includes PBKDF2-SHA1 and AES128-ECB cipher. The application this implementation password recovering verified ARM processor in Zynq SoC. In order compare performance against other platforms, we also algorithm on NVIDIA GeForce GTX 480 ATI Radeon R9 200 Series, multi-core CPU. Our superior advantage energy...
With the rapid development of Internet Things technology, smart speakers have become increasingly popular. However, speaker security is an ensuing threat. At present, are activated by voice, and they monitor users' voices 24 hours per day. Consequently, there may be problems with user privacy leakage. In this paper, we propose a non-intrusive digital forensic method for speakers. The main contribution paper effective combining network traffic analysis extraction intent alarms about abnormal...