A5102971264- Maritime Navigation and Safety
- Natural Language Processing Techniques
- Software Engineering Research
- Underwater Vehicles and Communication Systems
- Robotic Path Planning Algorithms
- Digital Media Forensic Detection
- Blockchain Technology Applications and Security
- Multimodal Machine Learning Applications
- Target Tracking and Data Fusion in Sensor Networks
- Adversarial Robustness in Machine Learning
- Interconnection Networks and Systems
- Advanced Malware Detection Techniques
- Generative Adversarial Networks and Image Synthesis
- Video Coding and Compression Technologies
- Speech and dialogue systems
- Handwritten Text Recognition Techniques
- Cloud Computing and Resource Management
- Embedded Systems Design Techniques
- Software Testing and Debugging Techniques
- Inertial Sensor and Navigation
- FinTech, Crowdfunding, Digital Finance
National University of Singapore
2024
Harbin Engineering University
2020-2021
Recent advances in large language models (LLMs) significantly boost their usage software engineering. However, training a well-performing LLM demands substantial workforce for data collection and annotation. Moreover, datasets may be proprietary or partially open, the process often requires costly GPU cluster. The intellectual property value of commercial LLMs makes them attractive targets imitation attacks, but creating an model with comparable parameters still incurs high costs. This...
Recent advances in large language models (LLMs) significantly boost their usage software engineering. However, training a well-performing LLM demands substantial workforce for data collection and annotation. Moreover, datasets may be proprietary or partially open, the process often requires costly GPU cluster. The intellectual property value of commercial LLMs makes them attractive targets imitation attacks, but creating an model with comparable parameters still incurs high costs. This...
In this paper, the ant colony algorithm used to study path planning for collision avoidance of unmanned underwater vehicle (UUV) in a large-scale marine environment. Multi-beam forward-looking sonar's simulation detection model based on grid environment established it. order overcome limitation that can only choose walk adjacent grids, concept virtual visualization introduced last step colony. This paper uses non-linear penalty mechanism objective function optimization avoid large-angle...
Path planning is the basic requirement for unmanned underwater vehicle (UUV) to complete tasks. According requirements of UUV path in working environment, a method based on QPSO algorithm proposed this paper. Firstly, obstacles environment are divided into known and unknown obstacles. For static an annular space model established global designed. fitness value each particle iterative process algorithm, optimization classification evolution proposed. Then, realtime information detected by...
State-of-the-art deep ISP models alleviate the dilemma of limited generalization capabilities across heterogeneous inputs by increasing size and complexity network, which inevitably leads to considerable growth in parameter counts FLOPs. To address this challenge, paper presents MetaISP - a streamlined model that achieves superior reconstruction quality adaptively modulating its parameters architecture response diverse inputs. Our rationale revolves around obtaining corresponding spatial...
With recent advancements in Large Multimodal Models (LMMs) across various domains, a novel prompting method called visual referring has emerged, showing significant potential enhancing human-computer interaction within multimodal systems. This offers more natural and flexible approach to human with these systems compared traditional text descriptions or coordinates. However, the categorization of remains undefined, its impact on performance LMMs yet be formally examined. In this study, we...
For capturing colored document images, e.g. posters and magazines, it is common that multiple degradations such as shadows, wrinkles, etc., are simultaneously introduced due to external factors. Restoring multi-degraded images a great challenge, yet overlooked, most existing algorithms focus on enhancing color-ignored via binarization. Thus, we propose DocStormer, novel algorithm designed restore documents their potential pristine PDF. The contributions are: firstly, "Perceive-then-Restore"...
In the field of underwater maneuvering target tracking, detection sensors is not effective in real time. Under man-made or non-man-made interference, observer only receives noise value abnormal value, that is, sensor will fail for a short Once fails to obtain correct data continuously period time, this fatal filter. Aiming at short-term failure sensors, paper proposes tracking method surface targets with compensation mechanism. This uses idea deep learning extract enough training samples...