1. 主動識別: 本研究團隊與普林斯頓大學及愛荷華州立大學合作,透過分析人機互動過程中使用者的生物特徵與行為特徵進行持續性且非干擾性身分識別。(CBS news: https://www.cnet.com/news/one-way-to-make-passwords-obsolete-just-keep-typing/)
2. 隱私維護機器學習: 將同態加密、多方安全計算、亂碼電路等加密機制加入機器學習演算法,以降低個人隱私資料在機器學習中遭到揭露的風險。
3. 核方法: 提升核方法的運算效率與強健性,以應用於大數據分析。
4. 深度學習理論與應用: 場景文字識別、影像除霧、姿態識別、全景視覺偵測、對抗例攻擊驗證。
1. Privacy Preserving Machine Learning: We study the embedding 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. Active Authentication: In collaboration with Princeton University and Iowa State University, we study how to continuously and unintrusively authenticate user's identity based on his/her behavior biometrics. (CBS news: https://www.cnet.com/news/one-way-to-make-passwords-obsolete-just-keep-typing/)
3. Kernel Methods: Cost-effectiveness and robustness issues in kernel-based machine learning.
4. Deep Learning Applications: Scene text recognition, image dehazing, pose estimation, 360 camera saliency detection, provable verification bound for adversarial example attack.