Page 85 - profile-ok
P. 85
研究人員 | Research Faculty
● Associate Research Fellow, IIS, Academia Sinica (1990 - )
張 復 Fu Chang ● MTS, Bell Laboratories, AT&T (1983 - 1990)
● Ph.D., Mathematical Statistics Columbia University (1979 - 1983)
● M.S., Mathematics, North Carolina State University (1976 - 1978)
● B.A., Philosophy, National Taiwan University (1969 - 1973)
副研究員 Associate Research Fellow
Ph.D., Mathematical Statistics, Columbia University
Tel: +886-2-2788-3799 ext. 1819
Fax: +886-2-2782-4814
Email: fchang@iis.sinica.edu.tw
http://www.iis.sinica.edu.tw/~fchang
代表著作 Publications
1. C.-H. Chou, W.-H. Lin, and F. Chang, A binarization method with
learning-built rules for document images produced by cameras, Pat-
研究簡介 Research Description tern Recognition, vol. 43, no. 4, pp. 1518-1530, 2010. (SCI , EI)
2. F. Chang and C.-H. Chou, A bi-prototype theory of facial attractive-
機器學習雖然已經廣受使用,在大型的問題上仍然有許 Although machine learning has been used extensively in various applications, it still ness, Neural Computation, vol. 21, no. 3, pp. 890-910, 2009. (SCI,
多困難。我們正在處理的大型問題有三種:大型的訓練 has difficulty in dealing with problems of large scales. There are three types of scales EI)
樣本,大型的類別數量,以及大型的(不相干的)特徵。 that one may face in applications: large scale in training samples, large scale in class 3. C.-H. Chou, W.-H. Lin, and F. Chang, Learning to binarize document
rd
針對第一種問題,我們提出了極有效率的決策樹分割方 types, and large scale in (irrelevant) features. For the first problem, we have pro- images, 3 International Workshop on Camera-Based Document
Analysis and Recognition, Barcelona, 2009.
法來訓練非線性支持向量機,可以提升訓練速度達數百 posed an extremely efficient tree decomposition approach to train non-linear sup-
倍,甚至數千倍,而依然維持相當的測試正確率。我們 port vector machines at a speedup factor of hundreds, sometimes even thousands, 4. F. Chang, C.-Y. Guo, X.-R. Lin, C.-J. Lu, Tree decomposition for
SVMs: experimental and theoretical results, Technical Report,
將此方法有效地應用於蛋白質與蛋白質介面的預測問 while achieving comparable test accuracy. This method has been used effectively to number TR-IIS-09-002, Institute of Information Science, Academia
題,得到300倍的加速效應。決策樹分割方法可以推廣 deal with a large size protein-protein interface prediction problem with a 300-fold Sinica, 2009.
到同樣有效力的森林分割方法,可使用於大型數量以 speedup. The tree decomposition method can be extended to an equally powerful 5. F. Chang and J.-C. Chen, An adaptive multiple feature subset method
及大型類別的資料上,以便同時解決第一與第二種問 forest decomposition in order to speed up machine learning on data sets that scale for feature ranking and feature selection, Technical Report, Number
題。針對第三種問題,我們正在實驗一種新方法,使用 up in both training samples and class types, thereby solving the first and the second TR-IIS-09-010, Institute of Information Science, Academia Sinica,
多重的特徵子集合來排列以及選取特徵,可以加快計算 problems simultaneously. For the third problem, we are pioneering a new method 2009.
速度,提升測試正確率,增加排列在不相干特徵之前的 for ranking and selecting features using mul-tiple feature subsets, and have gained 6. C.-C. Wu, C.-H. Chou, and F. Chang, A machine-learning approach
關鍵特徵數量,以及被選取的特徵所包含的關鍵特徵的 advantages in computing speed, test accuracy, the number of essential features that for analyzing document layout structures with two reading orders,
數量。除了發展新方法,我們也將所撰寫的軟體以及所 are ranked above all irrelevant features, and the number of essential features in the Pattern Recognition, vol. 41, pp. 3200-3213, 2008. (SCI, EI)
製造的資料公布於網路上,以方便未來可能的使用者。 selected features. While endeavoring to develop new methods, we also publicize 7. F. Chang, Techniques for solving the large-scale classification prob-
both our implementations and the data sets that were created in our applications, lem in Chinese handwriting recognition, Arabic and Chinese Hand-
writing Recognition, Lecture Notes in Computer Science, Series No.
so as to benefit potential users of our methods. 4768, 2008. (EI)
8. Y.-H. Liu, C.-C. Lin, W.-H. Lin, and F. Chang, Accelerating feature-
vector matching using multiple-tree and sub-vector methods, Pattern
Recognition, vol. 40, no. 9, pp. 2392-2399, 2007. (SCI, EI)
9. C.-H. Chou, S.-Y. Chu, and F. Chang, Estimation of skew angles for
scanned documents based on piecewise covering by parallelograms,
Pattern Recognition, vol. 40, no. 2, pp. 443-455, 2007. (SCI, EI)
10. C.-H. Chou, C.-Y. Guo, and F. Chang, Recognition of fragmented
characters using multiple feature-subset classifiers, Inter. Conf. Docu-
ment Analysis and Recognition, Brazil, 2007. (EI)
11. F. Chang, C.-C. Lin, and C.-J. Lu, Adaptive prototype learning al-
gorithms: theoretical and experimental studies, Journal of Machine
Learning Research, vol. 7, pp. 2125-2148, 2006. (SCI)
12. C.-H. Chou, C.-C. Lin, Y.-H. Liu, and F. Chang, A prototype classi-
fication method and its use in a hybrid solution for multiclass pattern
recognition, Pattern Recognition, vol. 39, no. 4, pp. 624-634, 2006.
(SCI, EI)
13. C.-H. Chou, B.-H. Kuo, and F. Chang, The generalized condensed
nearest neighbor rule as a data reduction method, Intern. Conf. Pat-
tern Recognition, Hong Kong, 2006. (EI)
84 85