A5032157872- Computer Graphics and Visualization Techniques
- 3D Shape Modeling and Analysis
- Advanced Vision and Imaging
- Natural Language Processing Techniques
- Video Analysis and Summarization
- Advanced Steganography and Watermarking Techniques
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Advanced Graph Neural Networks
- Generative Adversarial Networks and Image Synthesis
- Advanced Numerical Analysis Techniques
- Handwritten Text Recognition Techniques
- Digital Media Forensic Detection
Yunnan University of Finance And Economics
2021-2023
Large Language Models (LLMs) exhibit strong general-purpose language capabilities. However, fine-tuning these models on domain-specific tasks often leads to catastrophic forgetting, where the model overwrites or loses essential knowledge acquired during pretraining. This phenomenon significantly limits broader applicability of LLMs. To address this challenge, we propose a novel approach compute element-wise importance parameters crucial for preserving general fine-tuning. Our method utilizes...
Although reversible data hiding technology is widely used, it still faces several challenges and issues. These include ensuring the security reliability of embedded secret data, improving embedding capacity, maintaining quality media data. Additionally, irregular types, such as three-dimensional point clouds triangle mesh-represented 3D models, lack an ordered structure in their representation. As a result, these into digital does not provide sufficient information for complete recovery...
In general, the animation between three-dimensional models uses linear interpolation to calculate intermediate state. Before interpolation, mapping relationship source model and target should be calculated find one-to-one correspondence of vertices. It is often necessary traverse all vertices mesh in calculation, it can difficult implement parallel operation due irregularity triangular mesh. This paper aimed form a regular representation 3D conformal parameterization set up efficient...
Conventional Multi-Task Learning (MTL) models, such as hard sharing, adopt handcrafted network architecture, which shares entire layers for all tasks, and thus have two shortcomings: 1) negative transfer phenomenon 2) low parameter efficiency. This paper proposes a novel neural model, allows different tasks to share at the level. Specifically, model defines subnet each task by adopting task-specific binary masks. The masks are trainable can be learned together with weights using standard...