黃仁竑

ADAPT: Inference-Time Backdoor Detection via Representation-Disjoint Multi-Subspace Fusion for Hyperspectral Image Classifiers

智慧計算與科技研究所

高光譜影像分類、後門攻擊、推論階段偵測、無洩漏偵測、多子空間融合、光譜空間轉換器、hyperspectral image classification、backdoor attack、inference-time detection、no-leakage detection、multi-subspace fusion、spectral–spatial transformer

Extracting Attack Path Based on Multiple Sightings Using Graph-based Algorithms

智慧計算與科技研究所智慧物聯網產業碩士專班

攻擊路徑重建、多階段攻擊關聯、資安鑑識、異質安全資料、動態決策融合、加權攻擊圖、反向束搜尋、attack path reconstruction、multi-step attack correlation、cybersecurity forensics、heterogeneous security data、Dynamic Decision Fusion、weighted attack graph、Reverse Beam Search

Evaluating Indirect Prompt Injection(IPI) on LLM-Based Agentic Systems

智慧計算與科技研究所

間接提示注入、代理、大型語言模型、指令、漏洞、Indirect Prompt Injection、LLM Agents、Agentic AI Security、Adversarial Robustness、Prompt Injection Defense

Early Prediction of Intradialytic Hypotension Using Routine Dialysis Data: A Patient-Level Evaluation and Feature Source Analysis

智慧計算與科技研究所

血液透析、透析中低血壓、早期預測、機器學習、LightGBM、病人層級驗證、特徵來源分析、hemodialysis、intradialytic hypotension、early prediction、machine learning、patient-level validation、feature source analysis

Two-Stage Semi-Supervised Federated Continual Learning for Multi-Class Network Intrusion Detection

智慧計算與科技研究所

網路入侵偵測、聯邦學習、持續學習、半監督學習、偽標籤、Network intrusion detection、Federated learning、Continual learning、Semi-supervised learning、Pseudo labeling

Active Defense System for SSH Honeypot Based on Hierarchical Deep Reinforcement Learning and the MITRE Engage Framework

智慧系統與應用研究所

主動式防禦、蜜罐、SSH、階層式深度強化學習、MITRE Engage、欺敵防禦、Active Defense、Honeypot、Hierarchical Deep Reinforcement Learning、Cyber Deception

Bipartite Class-Band Graph Convolutional Network for Stealthy Universal Backdoor Attacks on Hyperspectral Image Classification

智慧計算與科技研究所智慧物聯網產業碩士專班

後門攻擊、高光譜影像分類、圖卷積網路、二部圖、觸發器生成、光譜波段選擇、遙測安全、Backdoor attack、hyperspectral image classification、graph convolutional network、bipartite graph、trigger generation、spectral band selection、remote sensing security

TrustCLIP: Mitigating Backdoors When Fine Tuning Pre‑Trained CLIP Models

智慧計算與科技研究所

後門攻擊、多模態對比學習、相似度分析、機器反學習、圖像分類、Backdoor Attack、Multimodal Contrastive Learning、Similarity Analysis、Machine Unlearning、Image Classification

GPT4MQTT:Vulnerability Detection via Automatic Extraction of Security Test Points with Self-Consistency Semantic Filtering and Domain-Aware Prompting

智慧與綠能產學研究所

MQTT、協定安全測試、大型語言模型、提示工程、自我一致性、行為層級漏洞、軟體測試、security testing、large language models (LLMs)、prompt engineering、self-consistency、vulnerability detection、few-shot learning

Inference-Time Detection of Backdoor Inputs in Untrusted CNN Models

智慧計算與科技研究所

後門攻擊、卷積神經網路、激活值分析、推論階段檢測、圖像分類、Backdoor Attack、Convolutional neural network、Activation analysis、Inference-time detection、Image classification

Imperceptible Universal Backdoor Attack

智慧計算與科技研究所

電腦視覺、後門攻擊、卷積神經網路、資訊安全、圖卷積神經網路、computer vision、backdoor attack、convolutional neural network、information security、graph convolutional neural network

Reinforcement Learning-based High-Interaction AD Honeypot Aligned with MITRE Engage

智慧系統與應用研究所

主動式防禦、蜜罐、強化學習、MITRE Engage、Active Directory、Active Defense、Honeypot、Reinforcement Learning

MITRE Engage-Aware Generative Threat Variants for Active Directory Deception Deployment Assessment

智慧計算與科技研究所

Active Directory、網路欺敵、生成式威脅變異、MITRE Engage、部署評估、大型語言模型、Agentic AI、檢索增強生成、Cyber Deception、Generative Threat Variants、Deployment Assessment、Large Language Model、Retrieval-Augmentation

Efficient MQTT Corpus Generation for IoT Fuzzer: An LLM-Based Approach

智慧系統與應用研究所

模糊測試、網路通訊協定、大語言模型、提示工程、Fuzzing、Network Prototol、Large Language Model、Prompt Engineering

A Relevance-based Data Selection Method for BERT-based Anomaly Detection System Using Unstructured Logs

智慧與綠能產學研究所

異常檢測、資料選擇、相關度、非結構化日誌、Anomaly detection、Data selection、Relevance、Unstructured logs

An Active Defense System Based on SSH Honeypot in Accordance with MITRE Engage

智慧系統與應用研究所

主動式防禦、蜜罐、欺騙技術、MITRE Engage、SSH、強化學習、威脅情報、Active defense、Honeypot、Deception techniques、Reinforcement Learning (RL)、Threat intelligence

Predict Peritoneal Dialysis Failure within the Next Three Months based on Deep Learning and Important Features Analysis

智慧與綠能產學研究所

腹膜透析、深度學習模型、時序資料、非監督式學習、資料前處理、peritoneal dialysis、preprocess、unsupervised、time series、deep learning