A5100658801- Speech and Audio Processing
- Music and Audio Processing
- Face recognition and analysis
- Biometric Identification and Security
- EEG and Brain-Computer Interfaces
- Video Coding and Compression Technologies
- Blind Source Separation Techniques
- Hedgehog Signaling Pathway Studies
- Music Technology and Sound Studies
- Speech Recognition and Synthesis
- Advanced Steganography and Watermarking Techniques
- Photoreceptor and optogenetics research
- Outsourcing and Supply Chain Management
- Engineering Applied Research
- Reconstructive Facial Surgery Techniques
- Chaos control and synchronization
- User Authentication and Security Systems
- Satellite Communication Systems
- Advanced Wireless Communication Techniques
- Collaboration in agile enterprises
- Migraine and Headache Studies
- Optimization and Search Problems
- PAPR reduction in OFDM
- Spectroscopy and Chemometric Analyses
- Underwater Vehicles and Communication Systems
Sun Yat-sen University
2007-2025
Sun Yat-sen Memorial Hospital
2021-2025
Language Science (South Korea)
2025
China Academy of Space Technology
2017-2024
Shanghai Jiao Tong University
2023-2024
Kuaishou (China)
2022
University of Electronic Science and Technology of China
2021
Jingdong (China)
2020
Chinese Academy of Sciences
2013
University of Chinese Academy of Sciences
2013
With diverse presentation attacks emerging continually, generalizable face anti-spoofing (FAS) has drawn growing attention. Most existing methods implement domain generalization (DG) on the complete representations. However, different image statistics may have unique properties for FAS tasks. In this work, we separate representation into content and style ones. A novel Shuffled Style Assembly Network (SSAN) is proposed to extract reassemble features a stylized feature space. Then, obtain...
Electroencephalography (EEG) is a vital noninvasive technique used in neuroscience research and clinical diagnosis. However, EEG data have complex nonEuclidean structure are often scarce, making training effective graph neural network (GNN) models difficult. We propose "pre-train, prompt" framework networks for analysis, called GNN-based Prompt Learning (GEPL). The first uses unsupervised contrastive learning to pre-train on large-scale dataset. It then transfers the generic knowledge...
Abstract This research examines how congenital visual or hearing impairment reshapes brain function using EEG. The study involved 40 children with impairment, and 42 age gender-matched normal as controls. investigation included assessments of auditory abilities, along comprehensive EEG evaluations. Techniques such source localization, functional connectivity cross-frequency coupling were used to analyse variations in activity. Machine learning methods, specifically support vector machines,...
Audio-Visual Speech Recognition (AVSR) is a promising approach to improving the accuracy and robustness of speech recognition systems with assistance visual cues in challenging acoustic environments. In this paper, we present novel audio-visual architecture unified cross-modal attention. Our concatenates sequences temporally from different modalities encodes fused sequence feature space using shared Conformer encoder. We then explicitly model additive noise potential out-of-sync samples...
Audio-visual speech recognition (AVSR) takes advantage of noise-invariant visual information to improve the robustness automatic (ASR) systems. While previous works mainly focused on clean condition, we believe modality is more effective in noisy environments. The challenges arise from difficulty adaptive fusion audio-visual and possible interferences inside training data. In this paper, present a new model with unified cross-modal attention mechanism. particular, auxiliary evidence combined...
Electroencephalogram (EEG) is an important technology to explore the central nervous mechanism of tinnitus. However, it hard obtain consistent results in many previous studies for high heterogeneity In order identify tinnitus and provide theoretical guidance diagnosis treatment, we propose a robust, data-efficient multi-task learning framework called Multi-band EEG Contrastive Representation Learning (MECRL). this study, collect resting-state data from 187 patients 80 healthy subjects...
In the study of two-sided matching decision problems, preference ordinal information is a key factor. However, in real life, it often difficult to ascertain complete information, and most cases we can only obtain an interval-valued information. this paper, strict based on multi-attribute discussed. As generalised model, adequately considers requirement satisfaction degree agents. Firstly, ranking method probability introduced deal with various interval numbers. Then, case multiple...
Discovering peaks is a premise of audio fingerprinting algorithms for information retrieval, which focus on spectrogram peak pairs. In this paper, we discuss finding in two-direction scanning method use dynamic threshold vector to find local maximal point as peaks. And then an improved using slide window with two grids proposed discovering at real-time. method, key steps are executed alternately: and sliding. each step, new found. When the used, experiments database 400,000 songs show that...
Objective: Vestibular migraine (VM) is one of the most common causes recurrent vertigo, but neural mechanisms that mediate such symptoms remain unknown. Since visual and photophobia are clinical features VM patients, we hypothesized patients have abnormally sensitive low-level processing capabilities. This study aimed to investigate cortex abnormalities in using evoked potential (VEP) standardized low-resolution brain electromagnetic tomography (sLORETA) analysis. Methods: We employed...
Video security is one of multi-view video applications. However the current coding (MVC) software does not contain any rate control technique, this paper proposes a algorithm for MVC based on quadratic rate-distortion (R-D) model. The proposed adopts fluid-flow traffic model, HRD (hypothetical reference decoder) and linear prediction model MAD (the mean absolute difference). Compared to with fixed quantization parameter, scheme can efficiently bit an average error 0.62% while keeping high efficiency.
As a mainstream means of satellite communication anti-jamming, spread spectrum (SSC) is faced with immense technical bottleneck, which the insufficient interference immunity, to break through in recent years. A novel Blind Source Separation (BSS) method proposed solve problem this paper. Considering statistical independence SSC signal and common jamming, BSS applied into receiver end anti-interference system parallel FastICA algorithm based on negentropy maximization utlized guide separating...
This paper proposes an improved QPSK modulation scheme which is called special QPSK. The key aspect of that the synchronous sequence hidden in information bits and does not occupy spectrum resource. In order to implement frame synchronization, correlation acquisition has been considered. Based on this approach, we derive closed-form expression expected peak values for cases signals aligned or aligned. Then constellation proposed optimized by minimizing bit error rate with constraint...
Numerous task-based functional magnetic resonance imaging studies indicate the presence of compensatory improvement in patients with congenital cataracts. However, there is neuroimaging evidence that shows decreased sensory perception or cognition information processing related to visual dysfunction, which favors a general loss hypothesis. This study explored connectivity between and other networks children cataracts using resting state electroencephalography. Twenty-one (age: 8.02 ± 2.03...
Face anti-spoofing (FAS) plays a crucial role in securing face recognition systems. Empirically, given an image, model with more consistent output on different views of this image usually performs better, as shown Fig.1. Motivated by exciting observation, we conjecture that encouraging feature consistency may be promising way to boost FAS models. In paper, explore thoroughly enhancing both Embedding-level and Prediction-level Consistency Regularization (EPCR) FAS. Specifically, at the...
With diverse presentation attacks emerging continually, generalizable face anti-spoofing (FAS) has drawn growing attention. Most existing methods implement domain generalization (DG) on the complete representations. However, different image statistics may have unique properties for FAS tasks. In this work, we separate representation into content and style ones. A novel Shuffled Style Assembly Network (SSAN) is proposed to extract reassemble features a stylized feature space. Then, obtain...