吳沛遠 Pei-Yuan Wu
主要研究領域:
機器學習,算法分析,估計理論,隱私強化,核方法Major Research Areas:
machine learning, algorithm analysis, estimation theory, privacy enhancing technologies, kernel method研究領域摘要:
-
隱私保護機器學習:
(1) 將密碼學方法（例如同態加密、安全多方計算和混淆電路）應用於機器學習算法，以降低洩露機密個人信息的風險。
(2) 支援密碼學的深度學習模型。
(3) 設計具備魯棒性(robustness)的機器學習算法。 -
機器學習算法理論分析:
(1) 針對基於核方法的強化學習的遺憾界(regret bound)分析。
(2) 在對抗性環境下，分析學習算法的魯棒性。
(3) 深度學習模型在對抗性環境下的可驗證性界限。
(4) 大型語言模型（LLM）的理論基礎：透過數學分析構建理論框架，解釋LLM的能力，包括推理與算術能力。 -
深度學習的應用:
機器學習算法之跨領域應用（主要與電腦視覺、強化學習相關）。
最近更新：民國113年11月20日
Research Summary:
1. Privacy Preserving Machine Learning:
(1) Application of cryptography, such as homeomorphic encryption, secure multi-party computation, and garbled circuits, into machine learning algorithms, in order to reduce the risk of exposing confidential personal information.
(2) Cryptography-friendly deep learning model.
(3) Robust learning algorithm design.
2. Theoretical Analysis of Machine Learning Algorithms:
(1) Regret bound analysis for kernel-based reinforcement learning.
(2) Robustness analysis of learning algorithms under adversarial settings.
(3) Provable verification bound of deep learning models under adversarial settings.
(4) Theoretical foundations of Large Language Models (LLM): Developing theoretical frameworks to explain the capabilities of LLMs, including reasoning and arithmetic, through mathematical analysis.
3. Deep Learning Applications: Interdisciplinary research where machine learning algorithms are adopted to various applications (mostly computer vision/reinforcement learning related).
Last update: 2024-11-20
-
B.S.
National Taiwan University, 2009 -
M.S.
Princeton University, 2012 -
Ph.D.
Princeton University, 2015
-
Address
EE2-234,
Department of Electrical Engineering,
National Taiwan University,
Taipei 106, Taiwan -
Phone
+886-2-33664687 -
FAX
+886-2-23671909 -
Email:
