A5100703288- Machine Learning and Data Classification
- Advanced Neural Network Applications
- Distributed Control Multi-Agent Systems
- Speech and Audio Processing
- Stability and Control of Uncertain Systems
- Blind Source Separation Techniques
- Grey System Theory Applications
- Neural Networks Stability and Synchronization
- Advanced Adaptive Filtering Techniques
- Human Pose and Action Recognition
- Research studies in Vietnam
- Advanced Multi-Objective Optimization Algorithms
- Advanced Optical Sensing Technologies
- Business Strategies and Innovation
- Distributed and Parallel Computing Systems
- Ethics and Social Impacts of AI
- Advanced Memory and Neural Computing
- COVID-19 diagnosis using AI
- Consumer Behavior in Brand Consumption and Identification
- Big Data Technologies and Applications
- Multimodal Machine Learning Applications
- Artificial Intelligence in Healthcare and Education
- Wildlife Ecology and Conservation
- Stochastic Gradient Optimization Techniques
- Neuroscience and Neural Engineering
Lingnan Normal University
2025
Chevron (United States)
2023-2024
Shenzhen Institutes of Advanced Technology
2024
University of Chinese Academy of Sciences
2024
University of Electronic Science and Technology of China
2024
Guangdong University of Technology
2023-2024
Shanghai Dianji University
2023-2024
Inner Mongolia Electric Power Survey & Design Institute (China)
2024
Chongqing University of Posts and Telecommunications
2023
Pennsylvania State University
2023
Remote sensing (RS) change detection incurs a high cost because of false negatives, which are more costly than positives. Existing frameworks, struggling to improve the Precision metric reduce positive, still have limitations in focusing on interest, leads missed detections and discontinuity issues. This work tackles these issues by enhancing feature learning capabilities integrating frequency components information, with strategy incrementally boost Recall value. We propose an enhanced...
Human mobility is a fundamental aspect of social behavior, with broad applications in transportation, urban planning, and epidemic modeling. However, for decades new mathematical formulas to model phenomena have been scarce usually discovered by analogy physical processes, such as the gravity radiation model. These sporadic discoveries are often thought rely on intuition luck fitting empirical data. Here, we propose systematic approach that leverages symbolic regression automatically...
Remote sensing (RS) change detection incurs a high cost because of false negatives, which are more costly than positives. Existing frameworks, struggling to improve the Precision metric reduce positive, still have limitations in focusing on interest, leads missed detections and discontinuity issues. This work tackles these issues by enhancing feature learning capabilities integrating frequency components information, with strategy incrementally boost Recall value. We propose an enhanced...
Animal pose estimation and tracking (APT) is a fundamental task for detecting animal keypoints from sequence of video frames. Previous animal-related datasets focus either on or single-frame estimation, never both aspects. The lack APT hinders the development evaluation video-based methods, limiting real-world applications, e.g., understanding behavior in wildlife conservation. To fill this gap, we make first step propose APT-36K, i.e., large-scale benchmark tracking. Specifically, APT-36K...
In the rapidly advancing arena of large language models (LLMs), a key challenge is to enhance their capabilities amid looming shortage high-quality training data. Our study starts from an empirical strategy for light continual LLMs using original pre-training data sets, with specific focus on selective retention samples that incur moderately high losses. These are deemed informative and beneficial model refinement, contrasting highest-loss samples, which would be discarded due correlation...
Visual Geometry Group (VGG)-style ConvNet is an neural-network process units (NPU)-friendly network; however, the accuracy of this architecture cannot keep up with other well-designed network structures. Although some reparameterization methods are proposed to remedy weakness, their performance suffers from homogenization issue parallel branches, and preset shape convolution kernels also influences spatial perception. To address problem, we propose a diversity-learning (DL) block build...
Fairness has been a critical issue that affects the adoption of deep learning models in real practice. To improve model fairness, many existing methods have proposed and evaluated to be effective their own contexts. However, there is still no systematic evaluation among them for comprehensive comparison under same context, which makes it hard understand performance distinction them, hindering research progress practical them. fill this gap, paper endeavours conduct first large-scale...
Blind source separation under an underdetermined reverberation environment is a very challenging issue. The classical method based on the expectation maximization algorithm. However, it limited to high environments, resulting in bad or even invalid performance. To eliminate this restriction, room impulse response reshaping-based designed solve problem of blind signal reverberant environment. Firstly, reshaping technology influence audible echo environment, improving quality received signals....
As one of the most vital ecological regions in China, well-being Inner Mongolia section Yellow River Basin directly hinges upon comprehending variations its ecosystem. The current research puts emphasis on analysis single-factor indicators within Mongolian and lacks summarization regarding overall state ecosystem River. This study, using methods such as remote sensing interpretation model simulation, combined with ground surveys, analyzes macrostructure, quality status, service functions,...
In recent years, advancements in artificial intelligence, particularly the revolutionary applications of deep-learning-based computer vision and large language models, have significantly affected oil gas industry. These technological developments underscored growing importance diversity data analytics across various aspects industry’s daily operations. Maintaining this momentum, practitioners industry reexamined existing workflows realized substantial benefits that intelligence brings,...
Learning to optimize (L2O) has gained increasing popularity, which automates the design of optimizers by data-driven approaches. However, current L2O methods often suffer from poor generalization performance in at least two folds: (i) applying L2O-learned optimizer unseen optimizees, terms lowering their loss function values (optimizer generalization, or ``generalizable learning optimizers"); and (ii) test an optimizee (itself as a machine model), trained optimizer, accuracy over data...
Abstract Finite‐time containment control of second‐order nonlinear multi‐agent systems with disturbances is investigated. The network topology among agents assumed to be directed. An event‐triggered terminal sliding mode controller designed, which robust respect disturbances. Based on the strategy, accessibility manifold and consensus tracking obtained sliding‐mode dynamics are achieved in finite‐time. Continuous update avoided settling time system convergence evaluated. Moreover,...
Deep learning recommendation models (DLRMs) are used across many business-critical services at Facebook and the single largest AI application in terms of infrastructure demand its data-centers. In this paper we discuss SW/HW co-designed solution for high-performance distributed training large-scale DLRMs. We introduce a scalable software stack based on PyTorch pair it with new evolution Zion platform, namely ZionEX. demonstrate capability to train very large DLRMs up 12 Trillion parameters...
In the practical application of brain-machine interface technology, problem often faced is low information content and high noise neural signals collected by electrode difficulty decoding decoder, which makes it difficult for robotic to obtain stable instructions complete task. The idea based on principle cooperative shared control can be achieved extracting general motor commands from brain activity, while fine details movement hosted robot completion, or have control. This study proposes a...
A room impulse response reshaping-based underdetermined blind source separation algorithm is designed to solve the signal problem in a reverberant environment. The main contribution use of reshaping technology eliminate influence audible echo for environment, improving quality received signals. new mathematical model time-frequency mixing signals established reduce approximation error transformation caused by high reverberation. Furthermore, an improved expectation maximization proposed...
Learning to Optimize (L2O) has drawn increasing attention as it often remarkably accelerates the optimization procedure of complex tasks by ``overfitting" specific task type, leading enhanced performance compared analytical optimizers. Generally, L2O develops a parameterized method (i.e., ``optimizer") learning from solving sample problems. This data-driven yields that can efficiently solve problems similar those seen in training, is, same ``task distribution". However, such learned...
Throughout the present decade, oil and gas industry is experiencing enduring digital transformation, marked by continuous evolution. In pursuit of increased efficiency, companies have embraced cutting-edge technologies in artificial intelligence big data, effectively automating their existing work flows. Having undergone extensive exploration refinement, debate between physics-based models data-driven has now found a harmonious middle ground. This convergence paved way for significant...
Diffusion-based image synthesis has attracted extensive attention recently. In particular, ControlNet that uses image-based prompts exhibits powerful capability in tasks such as canny edge detection and generates images well aligned with these prompts. However, vanilla generally requires training of around 5000 steps to achieve a desirable control for single task. Recent context-learning approaches have improved its adaptability, but mainly edge-based tasks, rely on paired examples. Thus,...