魏澤人

Router-Based Layer-Wise Recomposition of Fixed LoRA Experts for Multi-Task Adaptation in Large Language Models

智慧計算與科技研究所

大型語言模型、LoRA、路由器、多任務適應、專家重組、Large Language Models、Router、Multi-Task Adaptation、Expert Recomposition

Semantic Disruption Attacks on Large Language Models via Invisible Character Injection

智慧系統與應用研究所

大型語言模型、不可見字元、字元級攻擊、AI 代理安全、Large Language Models、Invisible Unicode Characters、Character-Level Attacks、AI Agent Security

Automated Prompt Optimization for Intervention Timing and Response Strategy Control in Tool-Use Multi-Party Conversational Agents

智慧系統與應用研究所・共同指導：郭志義

多方對話代理人、雙層代理人架構、自動化提示最佳化、競技場評估框架、遺傳帕雷托提示演化、Multi-Party Conversational Agent、Dual-Layer Agent Architecture、Automated Prompt Optimization、Arena Evaluation Framework、Genetic-Pareto Prompt Evolution

Improving GUI Grounding Performance with Lightweight Vision-Language Models

智慧計算與科技研究所

GUI 定位、視覺定位、視覺語言模型、輕量模型、少量資料訓練、GUI Grounding、Visual Grounding、Vision-Language Models、Lightweight Models、Low-Resource Training

Efficient GRPO Training Strategy for Qwen2.5 3B on Reasoning Tasks with Limited GPU Resources

智慧計算與科技研究所

大型語言模型、GRPO、強化學習、推理任務、資源受限訓練、格式化微調、精煉模型、Large Language Models、Reinforcement Learning、Reasoning Tasks、Resource-Constrained Training、Format-Aligned Pretraining、Distillation

Improving Reasoning About Differences in Image Sets in Vision-Language Models via Chain-of-Thought Prompting

智慧計算與科技研究所

圖像差異、視覺語言模型、Chain-of-Thought、提示學習、Qwen2.5-VL、Set Difference Captioning、Vision-Language Model、Chain-of-Thought Prompting、Visual Reasoning

Integration of Diffusion Models and Object Detection for Defect Detection in Panels

智慧與綠能產學研究所

擴散模型、圖片重建、瑕疵檢測、瑕疵分類、特徵融合、Diffusion Model、Inpainting、Anomaly detection、Defect Classification、Feature Fusion

ControlNet Aligner: A ControlNet-Based Subject-Driven Text-to-Image Generation Model

智慧系統與應用研究所

擴散模型、Diffusion model、FastComposer、ControlNet

Improving Unsupervised Visual Anomaly Detection Via Pretrained Stable Diffusion Models

智慧系統與應用研究所

異常檢測、非監督式學習、影像重建、擴散模型、穩定擴散模型、Anomaly Detection、Unsupervised Learning、Image Reconstruction、Diffusion Models、Stable Diffusion Models、DreamBooth

Features Fusion Decision System for Limited Industrial Panel Defect Classification

智慧計算與科技研究所

深度學習、瑕疵分類、少量樣本、特徵工程、特徵融合、Defect Classification、Limited Data set、Features Engineering、Features Fusion、Deep Learning

Segmentation using Lightweight Feature Extractor and Diffusion Model

智慧與綠能產學研究所

擴散模型、影像分割、Diffusion Model、Segmentation

Multi-Class Anomaly Detection with a Unified Diffusion Model

智慧計算與科技研究所

瑕疵檢測、擴散模型、圖片重建、Anomaly detection、Diffusion Model、Inpainting

Multi-Modal Image Registration with Edge Feature Filtering for Enhanced Alignment

智慧與綠能產學研究所

熱影像、可見光影像、影像校正、空間變換、電腦視覺、thermal image、visible image、image registration、spatial transformation、computer vision

Traffic Sign Detection Using Object Detection and Few-shot Classification

智慧計算與科技研究所

交通號誌、元學習、交通號誌資料集、物件偵測系統、少樣本學習、traffic sign、meta-learning、object detection system、few-shot learning、meta-learning dataset

Detecting Smart Contract Security Vulnerabilities with Natural Language Processing Method

智慧與綠能產學研究所

自然語言處理、區塊鏈、智能合約、特徵學習、Natural language Processing、Smart Contract、Blockchain、Representation Learning

One-Shot Voice Conversion Using Generative Adversarial Networks and Modulated Convolution

智慧與綠能產學研究所

單樣本、自動編碼器、語音轉換、生成對抗網路、特徵分離、one-shot、autoencoder、voice conversion、generative adversarial network、disentangled representations

User-Guided Line Art Colorization with StyleGAN

智慧計算與科技研究所

生成對抗網路、生成模型、自動編碼器、自動上色、圖對圖轉換、Generative Adversarial Network、Generative Model、AutoEncoder、Automatic Colorization、Image-to-Image translation

Exploration for Improving Emotion-based Symbolic-domain Music Recognition and Generation

智慧系統與應用研究所

音樂資訊檢索、符號音樂、音樂情緒、音樂生成、音樂分類、音樂情緒分類、基於情緒音樂生成、music information retrieval、symbolic music、music emotions、music generation、music classification、emotion music classification、emotion music generation

Face Swapping Improvement on Few-shot Learning Image-to-image Style Transfer

智慧系統與應用研究所

深度學習、電腦視覺、非監督式學習、風格轉換、換臉效果、影像去背、Deep learning、Computer vision、Unsupervised learning、Style transfer、Face swapping、Background removal

Improve Image Quality of StyleGAN via Knowledge Distillation

智慧系統與應用研究所

圖像生成、生成對抗網路、知識蒸餾、資料擴增、Image synthesis、GAN、Knowledge distillation、Data augmentation