A5100330018- Topic Modeling
- Natural Language Processing Techniques
- Multimodal Machine Learning Applications
- Advanced Text Analysis Techniques
- Speech Recognition and Synthesis
- Biometric Identification and Security
- Solar and Space Plasma Dynamics
- Speech and Audio Processing
- Complex Network Analysis Techniques
- Solar Radiation and Photovoltaics
- Text Readability and Simplification
- Domain Adaptation and Few-Shot Learning
- Rough Sets and Fuzzy Logic
- Human Mobility and Location-Based Analysis
- Educational Technology and Assessment
- Neural Networks and Applications
- Digital Communication and Language
- Face recognition and analysis
- Gout, Hyperuricemia, Uric Acid
- Pharmacological Effects of Natural Compounds
- Spam and Phishing Detection
- Urticaria and Related Conditions
- Advanced Authentication Protocols Security
- Electronic Health Records Systems
- Robotics and Sensor-Based Localization
Soochow University
2022-2024
Kunming University of Science and Technology
2012-2023
Tencent (China)
2022
RMIT University
2022
Northeast Normal University
2019-2020
In recent years, personality has been regarded as a valuable personal factor being incorporated into numerous tasks such sentiment analysis and product recommendation. This led to widespread attention text-based recognition task, which aims identify an individual's based on given text. Considering that ChatGPT recently exhibited remarkable abilities various natural language processing tasks, we provide preliminary evaluation of task for generating effective data. Concretely, employ variety...
Abstract To obtain a speaker’s pronunciation characteristics, method is proposed based on an idea from bionics, which uses spectrogram statistics to achieve characteristic give stable representation of the linear superposition short-time spectrograms. deal with issue slow network training and recognition speed for speaker systems resource-constrained devices, traditional SOM neural network, adaptive clustering self-organizing feature map (AC-SOM) algorithm proposed. This automatically...
As the cost associated with fine-tuning Large Language Models (LLMs) continues to rise, recent research efforts have pivoted towards developing methodologies edit implicit knowledge embedded within LLMs. Yet, there's still a dark cloud lingering overhead -- will editing trigger butterfly effect? since it is unclear whether might introduce side effects that pose potential risks or not. This paper pioneers investigation into pitfalls for To achieve this, we new benchmark datasets and propose...
Situation-awareness (SA) has been important for natural disaster management and smart decision making. Traditionally, security officers recognize situations through emergency reporting with phone calls. However, due to the busy lines or power outages caused by disasters, traditional SA limited in terms of time mitigation response, which may cause high loss life properties areas. Social media-based studied recently. existing systems are designed events without nonconsecutive migrations over...
We propose PolyVoice, a language model-based framework for speech-to-speech translation (S2ST) system. Our consists of two models: model and speech synthesis model. use discretized units, which are generated in fully unsupervised way, thus our can be used unwritten languages. For the part, we adopt existing VALL-E X approach build unit-based audio This grants ability to preserve voice characteristics speaking style original speech. examine system on Chinese $\rightarrow$ English Spanish...
Complex social event summarization is a problem which has been shown having great utility for real-world applications, including crisis management, rumor control and government policy tracking. In recent years there significant research effort spent on effectively extracting meaningful textual descriptions of an event. However, in many critical situations, events are complex context-sensitive, demands the online integrated manner. this paper, we propose first approach, namely SOMA,...
Abstract Solar images observed in different channels with instruments are crucial to the study of solar activity. However, have fields view, causing them be misaligned. It is essential accurately register for studying activity from multiple perspectives. Image registration described as an optimizing problem image registered a reference image. In this paper, we proposed novel coarse-to-fine method multichannel images. coarse step, used regular step gradient descent algorithm optimizer...
Objectives Gout, as the most prevalent form of inflammatory arthritis, necessitates use animal models to investigate molecular mechanisms involved in its development. Therefore, our objective was develop a novel chronic mouse model gout that more closely mimics progression humans. Methods A established by simple method, which does not require high technical proficiency, predominantly involves daily intraperitoneal injections potassium oxonate for approximately 4 months, combined with...
Fine-tuning Large Language Models (LLMs) adapts a trained model to specific downstream tasks, significantly improving task-specific performance. Supervised Fine-Tuning (SFT) is common approach, where an LLM produce desired answers. However, LLMs with SFT sometimes make simple mistakes and result in hallucinations on reasoning tasks such as question-answering. Without external feedback, it difficult for learn good mapping between the question answer, especially small dataset. This paper...
Mobile dating applications like Bumble and Tinder have grown in popularity, increasingly attracting scholarly attention. Our study focuses on the impact of politeness strategies imposition conversational outcomes mobile apps. Using a 2 by factorial design we examine degree directness perceptions potential romantic partners attitudes toward intensifying relationship. We found that indirectness (a higher-order strategy) requests for face-to-face dates (high imposition) were positively...
Most state-of-the-art synthetic aperture radar (SAR) ship detection methods based on deep learning require large amounts of labeled data for network training. However, the annotation process requires significant manpower and resources especially SAR images, since relevant background knowledge should be necessary annotators. Considering available optical imagery datasets with labels, we propose an unsupervised oriented method cross-modality distribution alignment, termed as <monospace...
In this paper, we propose a matrix random low-rank approximation (MRLRA) approach to generate cancelable biometric templates for privacy-preserving. MRLRA constructs approximate the hybridization of feature and matrix. Theoretically analysis shows distance between one template by its original is very small, which results verification authentication performance near that templates. Cancelable conquer weakness projection based templates, in will deteriorate much under same tokens. Experiments...
Multi-hop inference for explanation generation is to combine two or more facts make an inference. The task focuses on generating explanations elementary science questions. In the task, relevance between and QA pairs of vital importance. To address a three-step framework proposed. Firstly, vector distance texts utilized recall top-K relevant each question, reducing calculation consumption. Then, selection module employed choose those most relative in autoregressive manner, giving preliminary...
Ring signature is one of most important type digital signature. Usually the ring generated by signer, however sometimes we need be subset from an access structure a group. Moreover, to overcome management problem private keys in traditional schemes. In this paper, novel biometric scheme for authorized subsets proposed. The members can cooperatively sign message based on their identity and verifier verify if structure, he cannot find out which has really issued analysis results show that...
BERT has shown remarkable performance in several natural language processing tasks, but it fails to exhibit the same high cross-lingual particularly machine translation. To address this issue, we propose a BERT-enhanced neural translation (BE-NMT) model that optimizes use of information contained by NMT. Our proposed comprises three components: (1) A MASKING strategy mitigate knowledge forgetting caused fine-tuning on NMT task; (2) Serial and parallel multi-attention models for incorporating...
This study proposes a technology for obtaining stable pronunciation features speaker recognition using statistical strategy that utilizes the physiological characteristics of human and bionic cognitive processes speaker's voice. Due to its reflecting frequency information in voice, spectrogram is employed analyze samples. An individual's voice signal first split into short-duration segments logarithmic then generated each segments. To collect energy at particular frequency, linear...
Although BERT has achieved excellent results in various natural language processing tasks, it does not exhibit the same high performance cross-lingual especially machine translation tasks. We propose a enhanced neural (BE-NMT) model to improve use of information that is contained by NMT. The consists three aspects: (1) A MASKING strategy applied alleviate knowledge forgetting caused fine-tuning on NMT task.(2) Serial and parallel are combined for multi-attention models when incorporating...
Although BERT has achieved excellent results in various natural language processing tasks, it does not exhibit the same high performance cross-lingual especially machine translation tasks. We propose a enhanced neural (BE-NMT) model to improve use of information that is contained by NMT. The consists three aspects: (1) A MASKING strategy applied alleviate knowledge forgetting caused fine-tuning on NMT task.(2) Serial and parallel are combined for multi-attention models when incorporating...