A5114500347- Topic Modeling
- Natural Language Processing Techniques
- EEG and Brain-Computer Interfaces
- Speech Recognition and Synthesis
- Software Engineering Research
- Stock Market Forecasting Methods
- Human Pose and Action Recognition
- Multimodal Machine Learning Applications
- Speech and Audio Processing
- Software Reliability and Analysis Research
- Neural Networks and Applications
- Time Series Analysis and Forecasting
- Neural dynamics and brain function
- Advanced Malware Detection Techniques
- Advanced Memory and Neural Computing
- Domain Adaptation and Few-Shot Learning
- Privacy-Preserving Technologies in Data
- Gait Recognition and Analysis
- Advanced Text Analysis Techniques
- Speech and dialogue systems
- Semantic Web and Ontologies
- Sentiment Analysis and Opinion Mining
- Text and Document Classification Technologies
- Imbalanced Data Classification Techniques
- Hand Gesture Recognition Systems
Taiyuan University of Science and Technology
2025
South China University of Technology
2024
Hertie School
2022
Kuaishou (China)
2022
Deakin University
2021
North China Electric Power University
2019
China Electric Power Research Institute
2019
Shandong Institute of Automation
2014
Prototypes help to explain the predictions of deep classification models for time series. However, most learn prototypes by randomly initializing an uncertain number low-discriminative prototypes, which may lead unstable and unreliable results. To address these issues, we propose a new class Discriminative Prototype Learning Network (DPL-Net), learns appropriate class-discriminative thus improving performance. Specifically, proposed Initialization Mechanism (PIM) introduces proximity metric...
Decoding auditory attention from brain activities, such as electroencephalography (EEG), sheds light on solving the machine cocktail party problem. However, effective representation of EEG signals remains a challenge. One reasons is that current feature extraction techniques have not fully exploited spatial information along signals. reflect collective dynamics activities across different regions. The intricate interactions among these channels, rather than individual channels alone,...
Auditory attention decoding (AAD) with electroencephalography (EEG) holds great promise in brain-computer interface (BCI). Despite much progress, it remains a research topic on how to effectively evaluate the performance of EEG-based AAD algorithms under an appropriate setting that reflects use scenarios. It is desired systems are evaluated cross-subject and cross-trial settings. However, often reported same-subject, same-trial settings, where test data not truly separated from training...
As we are aware, there millions of deaf-mutes around the world. It is a necessity to conduct research into sign language recognition as it massive significance helping normal people and deaf-mute communicate smoothly with others. A behavior method was proposed in this paper address issue. Inspired by Multi-Fiber Networks, CBAM-ResNet neural network that extended structure ResNet 3D convolution convolutional block attention module added. In fifth layer network, unit 3D-Res2Net used preserve...
In this work, we take a further step towards satisfying practical demands in Chinese lyric generation from musical short-video creators, respect of the challenges on songs' format constraints, creating specific lyrics open-ended inspiration inputs, and language rhyme grace. One representative detail these is to control at word level, that is, for songs, creators even expect fix-length words certain positions match special melody, while previous methods lack such ability. Although recent...
The hybrid speech synthesis system, which uses the acoustic model trained according to criterion of Maximum Likelihood select proper candidates from corpus, has become a hot topic in recent days. For this performance is affected by size base training unit and candidate unit. Most existed systems use same kind such as syllable or phone for both concatenation. In Mandarin, initials finals form fundamental elements pronunciation, are always chosen statistical parametric TTS system. paper new...
The real-world data sets often leveraged by Federated Learning (FL) applications are mostly non-independent and non-identically distributed (non-IID). This usually results from the diverse nature of participating clients their individual data-gathering contexts. An effective FL algorithm must incorporate capability to produce a joint model that generalizes captures these patterns. In this work, we show how using some wild external samples as placeholders for missing classes on client devices...
Natural language explanations (NLEs) are commonly used to provide plausible free-text of a model's reasoning about its predictions. However, recent work has questioned the faithfulness NLEs, as they may not accurately reflect internal process regarding predicted answer. In contrast, highlight -- input fragments identified critical for predictions exhibit measurable faithfulness, which been incrementally improved through existing research. Building on this foundation, we propose G-Tex,...
In terms of the security problem power information system, this paper analysed importance software defect prediction method in object-oriented development, and proposed a model based on particle swarm optimized Support Vector Machine (SVM) corresponding to features software. The mainly consists three parts: first is pre-processing module which normalizes original data selects feature, then second adaptive inertia weight optimizes parameters SVM with accuracy as fitness. Finally, last...
Current end-to-end autoregressive TTS systems (e.g. Tacotron 2) have outperformed traditional parallel approaches on the quality of synthesized speech. However, they introduce new problems at same time. Due to nature, time cost inference has be proportional length text, which pose a great challenge for online serving. On other hand, style synthetic speech becomes unstable and may change obviously among sentences. In this paper, we propose Phrase based Parallel End-to-End System (PPSpeech)...
The application of machine learning has played an important role in several aspects text classification across domains, and brought with it great changes to the current state art. In this paper, we propose a novel NLP techniques classify entities by their International Standard Industrial Classification (ISIC) code based on descriptions provided business owners themselves names said businesses. Faced issues irregularity small amount noisy training data, employ different models data...