A5066407787- Domain Adaptation and Few-Shot Learning
- Advanced Image and Video Retrieval Techniques
- Advanced Bandit Algorithms Research
- Advanced Neural Network Applications
- Industrial Vision Systems and Defect Detection
- Advanced Computational Techniques and Applications
- Machine Learning and Algorithms
- Image and Signal Denoising Methods
- Advanced Algorithms and Applications
- CCD and CMOS Imaging Sensors
- Ethics and Social Impacts of AI
- Cloud Computing and Resource Management
- Adversarial Robustness in Machine Learning
- Network Time Synchronization Technologies
- Augmented Reality Applications
- Image and Object Detection Techniques
- Robotics and Automated Systems
- Explainable Artificial Intelligence (XAI)
- Image Retrieval and Classification Techniques
- Advanced Decision-Making Techniques
- Embedded Systems and FPGA Design
- Embedded Systems Design Techniques
- Image Processing Techniques and Applications
- Image Processing and 3D Reconstruction
- Robotic Path Planning Algorithms
The University of Texas at Dallas
2022-2024
Inspur (China)
2022-2024
University of Jinan
2023
The endeavor to preserve the generalization of a fair and invariant classifier across domains, especially in presence distribution shifts, becomes significant intricate challenge machine learning. In response this challenge, numerous effective algorithms have been developed with focus on addressing problem fairness-aware domain generalization. These are designed navigate various types particular emphasis covariate dependence shifts. context, shift pertains changes marginal input features,...
The fairness-aware online learning framework has arisen as a powerful tool for the continual lifelong setting. goal learner is to sequentially learn new tasks where they come one after another over time and ensures statistic parity of coming task across different protected sub-populations (e.g. race gender). A major drawback existing methods that make heavy use i.i.d assumption data hence provide static regret analysis framework. However, low cannot imply good performance in changing...
In the problem of online learning for changing environments, data are sequentially received one after another over time, and their distribution assumptions may vary frequently. Although existing methods demonstrate effectiveness algorithms by providing a tight bound on either dynamic regret or adaptive regret, most them completely ignore with model fairness, defined as statistical parity across different sub-population (e.g., race gender). Another drawback is that when adapting to new...
The fairness-aware online learning framework has emerged as a potent tool within the context of continuous lifelong learning. In this scenario, learner’s objective is to progressively acquire new tasks they arrive over time, while also guaranteeing statistical parity among various protected sub-populations, such race and gender when it comes newly introduced tasks. A significant limitation current approaches lies in their heavy reliance on i.i.d (independent identically distributed)...
Aiming at the problems of low efficiency and poor real-time performance in printed circuit board (PCB) defect detection, a PCB detection method based on improved YOLOv5 is proposed, which integrates module multiscale attention mechanism multi-branch. A shallow layer added to detect smaller targets fused with features deep network. An optimized anchor clustering was used obtain more suitable size for dataset. The Convolutional Block Attention Module (CBAM) introduced reweight assign important...
The fairness-aware online learning framework has emerged as a potent tool within the context of continuous lifelong learning. In this scenario, learner's objective is to progressively acquire new tasks they arrive over time, while also guaranteeing statistical parity among various protected sub-populations, such race and gender, when it comes newly introduced tasks. A significant limitation current approaches lies in their heavy reliance on i.i.d (independent identically distributed)...
Abstract The stability of the entire base station depends on synchronization station’s clock. clock management chip 8A34002 supports SyncE Ethernet and IEEE 1588. In this architecture, GPS receiver, SSI, master/slave switching device all emit PPS/TOD signals. signals are input into FPGA, which outputs one signal. signal enters in design scheme, DPLL provides filtering clock-following output. 8A34002‘s output is used as system’s after processing. chip, X86 main control BBU board driven by...
Transformer-based models have been verified successful in many natural language processing and computer vision tasks. Because of computational complexity, efficient transformer variants proposed, including the Long-former, which aims for long document processing. In this paper, we present an effective post-training quantization scheme Longformer. Based on sliding window attention Longformer, propose chunkwise quantization. It can decrease noise caused by significant gaps between ranges...
Currently, the robot system has been widely applied and rapidly developed. It applies to diverse fields such as service, patrol, life, etc. Most patrol systems are based on augmented reality SLAM complete guided navigation tasks. The AI is embedded in for objects detection during Since traditional model computationally intensive at deployment, it requests special computing devices support, GPUs TPUs. However, edge have too limited memory run models, which significantly limits application of...
The rapid development of artificial intelligence and big data has put forward a higher requirements for database performance. As one the basic operations database, sorting based on CPU operation cannot meet requirements. Therefore, heterogeneous method combining FPGA been developed. However, these methods only sort fixed format data, are not optimized storage so it is difficult to apply directly. Aiming at above problems, this paper proposes an acceleration table parsing mask calculation....