謝秉均

WhittleFormer: Scalable Neural Index Policies for Dynamic Counter-UAV Jamming

資訊科學與工程研究所

無休止多臂機問題、反無人機干擾、深度強化學習、Whittle 指數、Restless multi-armed bandits、Counter-UAV jamming、Deep reinforcement learning、Whittle index

From Reward-Free Representations to Preferences: Rethinking Offline Preference-Based Reinforcement Learning

資訊科學與工程研究所

離線強化學習、偏好式強化學習、零樣本強化學習、對比學習、無獎勵表徵學習、Offline Reinforcement Learning、Preference-based Reinforcement Learning、Zero-shot Reinforcement Learning、Contrastive Learning、Reward-Free Representation Learning

Learning to Optimize: Bandit Algorithms and Reinforcement Learning for Black-box Decision Problems

資訊科學與工程研究所

強化學習、多臂老虎機、貝葉斯最佳化、Reinforcement Learning、Bandits Learning、Bayesian optimization.

Action-Constrained Reinforcement Learning

資訊科學與工程研究所

強化學習、動作限制、多目標強化學習、策略最佳化、Reinforcement Learning、Action Constraints、Sampling、Multi-objective RL、Policy Optimization

Plan2Align: Predictive Planning Based Test-Time Preference Alignment for Large Language Models

數據科學與工程研究所

大型語言模型、測試階段對齊、模型預測控制、語言模型規劃、上下文學習、Large Language Models、Test-time Alignment、Model Predictive Control、LLM Planning、In-context Learning

Semi-Supervised Cross-Domain Imitation Learning

資訊科學與工程研究所

跨領域模仿學習、半監督式模仿學習、跨域映射、自適應加權、不完美示範、Cross-domain imitation learning、Semi-supervised imitation learning、Cross- domain mapping、Adaptive weighting、Imperfect demonstrations

Plan2Cleanse: Test-Time Backdoor Detection and Mitigation via Monte-Carlo Planning in Deep Reinforcement Learning

資訊科學與工程研究所

強化學習、後門偵測、後門緩解、測試階段防禦、蒙地卡羅樹搜尋、reinforcement learning、backdoor detection、backdoor mitigation、test-time defense、Monte Carlo tree search

A Reward-Free Viewpoint on Multi-Objective Reinforcement Learning

資訊科學與工程研究所

多目標強化學習、半監督式學習、無回饋強化學習、前向-後向表示法、跨目標 遷移、Multi-objective reinforcement learning、Semi-supervised learning、Reward-free reinforcement learning、Forward-backward representation、Cross-objective transfer

Bridging the Reality Gap in Reinforcement Learning: Robust Algorithms and Techniques for Practical Applications

資訊科學與工程研究所・共同指導：王昱舜

強化學習、穩健模型、現實落差、模擬環境、靜態離線數據集、Reinforcement Learning、Robust Model、reality gap、simulation、offline RL

Improving Anesthesia Simulators via Reinforcement Learning

資訊科學與工程研究所

麻醉、模擬器、序列模型、動態模型、強化學習、變分推斷、Anesthesia、simulator、sequence model、dynamics model、reinforcement learning、variational inference

Action-Constrained Imitation Learning

資訊科學與工程研究所

模仿學習、動作限制、軌跡最佳化、模型預測控制、動態時間規整、Imitation Learning、Action Constraint、Trajectory Optimization、Model Predictive Control、Dynamic Time Warping

A Study of Deep Reinforcement Learning for Human-Like Behavior Modeling and Scheduling Optimization

資訊科學與工程研究所・共同指導：吳毅成、王豐堅

深度強化學習、人類化行為、受限策略優化、擴增拉格朗日法、工作車間排程問題、彈性工作車間排程問題、動態彈性工作車間排程問題、圖神經網路、Deep Reinforcement Learning (DRL)、Human-Like Behavior、Constrained Policy Optimization、Augmented Lagrangian、Job Shop Scheduling Problem (JSP)、Flexible Shop Scheduling Problem (FJSP)、Dynamic Flexible Job Shop Scheduling Problem (DFJSP)、Graph Neural Networks (GNN)

Cross-Domain Knowledge Transfer for RL via Preference Consistency

智能系統研究所

強化學習、深度強化學習、跨領域遷移、遷移式學習、偏好式強化學習、基於人類回饋的強化學習、Reinforcement learning、Deep Reinforcement learning、Cross-domain transfer、Transfer learning、Preferencebased RL、Reinforcement Learning from Human Feedback

Survive on Planet Pandora: Robust Cross-Domain RL Under Distinct State-Action Representations

資訊科學與工程研究所

跨領域遷移、遷移學習、強化學習、領域間映射、負遷移、Cross-domain transfer、Transfer learning、Reinforcement learning、Inter-domain mapping、Negative transfer

Learning From Demonstrative Experts of Diverse Preference

網路工程研究所

逆強化學習、從示範數據學習、多專家、多目標、半監督式學習、Inverse reinforcement learning、learning from demonstrations、multiple experts、multi-objective learning、semi-supervised learning

Learning From Expert Demonstrations With Incomplete Observations

資訊科學與工程研究所

模仿學習、不完整觀察、透過演示學習、Imitation learning、Incomplete observations、Learning from demonstrations

Accelerated Policy Gradient: On the Convergence Rates of the Nesterov Momentum for Reinforcement Learning

智能系統研究所

強化學習、機器學習、最佳化、Nesterov 加速梯度、策略梯度、Reinforcement Learning、Machine Learning、Optimization、Nesterov's Accelerated Gradient、Policy Gradient

Provably Convergent Mixture-of-Experts in Reinforcement Learning

數據科學與工程研究所

混合專家模型、分層式強化學習、技能組合、漸進式收斂、收斂速率、迷宮、mixture-of-experts、hierarchical reinforcement learning、skill composition、asymptotic convergence、convergence rate、maze2d

Model Selection for Offline Model-Based RL: A Bayesian Optimization Approach

數據科學與工程研究所

離線強化學習、貝氏最佳化、動態模型選擇、動態模型距離、高斯過程、Offline Reinforcement Learning、Bayesian Optimization、Dynamics Model Selection、Dynamics Model Distance、Gaussian Process

Variance-Reduced Frank-Wolfe Policy Optimization For Action-Constrained Reinforcement Learning

數據科學與工程研究所

強化學習、行動約束強化學習、方差減少、Frank-Wolfe、Actor-Critic、Reinforcement Learning、Action-Constrained Reinforcement Learning、Variance-Reduction

Robustifying Proximal Policy Optimization Against Noisy Critics

資訊科學與工程研究所

強化學習、穩健性、策略最佳化、近端策略優化、鉸鏈策略優化、reinforcement learning、Policy optimization、Robustness、Proximal policy optimization、Hinge policy optimization、On-Policy

Multi-Objective Time-Varying Bayesian Optimization

數據科學與工程研究所

高斯過程、貝氏最佳化、多目標時變貝氏最佳化、Gaussian process、Bayesian optimization、Multi-objective time-varying Bayesian optimization

Adaptive-UCB for Online Restless Bandits

人工智慧技術與應用碩士學位學程

強化學習、多臂機學習、馬可夫鏈、置信區間上界、Reinforcement Learning、Bandits Learning、Markov Chain、Upper Confidence Bound

Coordinate Ascent Policy Optimization

資訊科學與工程研究所

策略梯度、座標向上、自然策略梯度、重要性採樣、離軌策略、Policy Gradient、Coordinate Ascent、Off-Policy、Natural Policy Gradient、Importance Sampling

Q-Pensive: Boosting Sample Efficiency of Multi-Objective RL Through Memory Sharing of Q-Snapshots

資訊科學與工程研究所

強化學習、多目標強化學習、多任務學習框架、Reinforcement Learning、Multi-Objective Reinforcement Learning、Multi-Task Learning、Soft Actor Critic

Accelerating Gaussian Process Regression via Meta-Learned Neural Processes: A Utility-Based MAML Approach 研究生：李昕（Lee, Xin） 指導教授：謝秉均（Hsieh, Ping-Chun） 中華民國一百一十一年二月 February 2022

資訊科學與工程研究所

高斯過程、高斯過程迴歸、元學習、模型無關元學習、小樣本學習、Gaussian Processes、Gaussian Process Regression、Meta learning、Model-agnostic meta learning、Few-shot learning

Reinforced Few-Shot Acquisition Function Learning for Bayesian Optimization

資訊科學與工程研究所

貝氏最佳化、強化學習、少樣本學習、機器學習、超參數最佳化、Bayesian optimization、Reinforcement learning、Few-shot learning、Machine learning、Hyperparameter optimization

Frank-­Wolfe Policy Optimization for Reinforcement Learning with Action Constraints

網路工程研究所

強化學習、監督式學習、動作限制強化學習、梯度消失、策略演算法、Reinforcement learning、Action­-constrained reinforcement learning、Policy optimization、Zero-­gradient、Supervised learning

On the Theory of Policy Improvement in Reinforcement Learning

應用數學系所・共同指導：吳毅成、吳金典

馬可夫決策過程、強化式學習、時間差分演算法、策略梯度、可解釋人工智慧、最佳化控制、自動控制、多旋翼機、Markov Decision Process、Reinforcement Learning、Temporal Difference Learning、Policy Gradient、Explainable AI、Optimal Control、Autonomous Control、Multicopter