Page 70 - profile2012.indd

Research Faculty 研究人員

副研究員 助研究員
張 復 Fu Chang 張原豪 Yuan-Hao Chang

Associate Research Fellow Assistant Research Fellow
Ph.D., Mathematical Statistics, Columbia University Ph.D., Computer Science and Information Engineering, National Taiwan
University
Tel: +886-2-2788-3799 ext. 1819 Fax: +886-2-2782-4814 Tel: 886-2-2788-3799 ext. 1612 Fax: 886-2-2782-4814
Email: fchang@iis.sinica.edu.tw Email: johnson@iis.sinica.edu.tw
http://www.iis.sinica.edu.tw/~fchang http://www.iis.sinica.edu.tw/pages/johnson

● Associate Research Fellow, IIS, Academia Sinica (1990 - ) ● M. S., Mathematics, North Carolina State University (1978) ● Master Thesis Supervision Award, Institute of Information and ● Ph. D. Paper Award of Taiwan Information Storage Association.
● Member of Technical Staff, Bell Laboratories, AT&T (1983 – 1990) ● B. A., Philosophy, National Taiwan University (1973) Computing Machinery (IICM). (2011) (2009)
● Ph. D., Mathematical Statistics, Columbia University (1983) ● Two-year’s project grant from National Science Council for excel- ● Best paper nomination at the 44th ACM/IEEE Design Automation
lent junior research investigators. (2011) Conference (DAC) (2007)

Research Description Publications Research Description Publications

My research interests lie in machine learning, document 1. F. Chang, C.-J. Chen, and C.-J. Lu, A linear-time component- My research interests lie in the area of computer systems, 1. Yuan-Hao Chang, Jen-Wei Hsieh, and Tei-Wei Kuo, “Endur-
image analysis, and some other related areas. With my lab labeling algorithm using contour tracing technique, Computer especially embedded systems, with the emphasis on stor- ance Enhancement of Flash-Memory Storage Systems: An Ef-

fellows, we created a few general methods as well as some Vision and Image Understanding, vol. 93, no. 2, pp. 206-220, age systems and operating systems. ficient Static Wear Leveling Design,” the ACM/IEEE Design
Automation Conference (DAC), San Diego, California , USA,
special techniques described as follows. 2004. One of my recent research focuses is on storage system Jun. 4-8, 2007. (Acceptance rate: 21%) (Best Paper Nomination
Among the general methods, our latest invention is a 2. F. Chang, C.-C. Lin, and C.-J. Lu, Adaptive prototype learn- designs for embedded systems. Embedded systems, es- (10/659) - Top Conference)
learning algorithm, called decision tree support vector ing algorithms: theoretical and experimental studies, Journal pecially battery-backed consumer electronics such as 2. Tei-Wei Kuo, Yuan-Hao Chang, Po-Chun Huang, and Che-Wei
machine (DTSVM), which can accelerate the training of of Machine Learning Research, vol. 7, pp. 2125-2148, 2006. smartphones, usually adopt ash memory as their stor- Chang, “Special Issues in Flash,” the ACM/IEEE International

SVMs by a factor of thousands and still achieves compa- 3. C.-H. Chou, C.-C. Lin, Y.-H. Liu, and F. Chang, A prototype age media. Because of the cost-reduction issue and the Conference on Computer-Aided Design (ICCAD), San Jose,
California, USA, Nov. 10-13, 2008. (Acceptance rate: 23%)

rable test accuracy. The other invention of ours is a linear- classification method and its use in a hybrid solution for mul- advance of manufacturing technologies, ash-based (Top Conference)

time component-labeling algorithm that identi es and ticlass pattern recognition, Pattern Recognition, vol. 39, no. 4, storage system designs face critical challenges on perfor- 3. Yuan-Hao Chang and Tei-Wei Kuo, “A Commitment-based
pp. 624-634, 2006.
labels connected components within a single pass over mance and reliability problems. Thus, we designed new Management Strategy for the Performance and Reliability En-

a 2D image and requires no re-labeling throughout the 4. Y.-H. Liu, C.-C. Lin, W.-H. Lin, and F. Chang, Accelerating cache systems for general-purpose le systems with new hancement of Flash-memory Storage Systems,” the ACM/IEEE
process. In developing the above two methods, we have feature-vector matching using multiple-tree and sub-vector non-volatile storage media to prevent from data losses Design Automation Conference (DAC), San Francisco, Jul. 26-
methods, Pattern Recognition, vol. 40, no. 9, pp. 2392-2399,
31, 2009. (Acceptance rate: 21%(148/682)) (Top Conference)

made available tool kits for free download by the general 2007. upon power losses and to improve the le-system’s e -

public. The log record shows that thousands of download 5. C.-C. Wu, C.-H. Chou, and F. Chang, A machine-learning ciency over ash storage devices; we also developed new 4. Yuan-Hao Chang, Jen-Wei Hsieh, and Tei-Wei Kuo, “Improv-
ing Flash Wear-Leveling by Proactively Moving Static Data,”
had been made from our website, re ecting the fact that approach for analyzing document layout structures with two management schemes to solve the problems inhibited in IEEE Transactions on Computers, vol. 59, no. 1, pp. 53-65, Jan.

the two methods have become popular instruments for reading orders, Pattern Recognition, vol. 41, pp. 3200-3213, the next-generation ash-memory chips of fast-growing 2010.
many researchers. The third invention of ours is a proto- 2008. capacity. Now we are exploring energy-saving technolo- 5. Yuan-Hao Chang, Jian-Hong Lin, Jen-Wei Hsieh, and Tei-Wei
type learning algorithm that has found applications in 6. F. Chang and C.-H. Chou, A bi-prototype theory of facial at- gies to realize the concept of “green storage” in embedded Kuo, “A Strategy to Emulate NOR Flash with NAND Flash,”
OCR and, surprisingly, also in predicting the attractiveness tractiveness, Neural Computation, vol. 21, no. 3, pp. 890-910, systems. ACM Transactions on Storage, vol. 6, no. 2, pp. 5:1-5:23, Jul.
2010.
of human faces. 2009. Another recent research focus is on the operating system
Among the techniques that aim for solving some special 7. F. Chang, C.-Y. Guo, X.-R. Lin, and C.-J. Lu, Tree Decom- designs for embedded systems. Due to the fast-growing 6. Yuan-Hao Chang and Tei-Wei Kuo, “A Management Strategy
for the Reliability and Performance Improvement of MLC-
problems, mainly oc-curring in the area of document im- position for Large-Scale SVM Problems, Journal of Machine capability of hardware resources and the fast-growing based Flash-memory Storage Systems,” IEEE Transactions on
age analysis, we have created a fast and reliable method Learning Research, 11(Oct), pp. 2935−2972, 2010. complexity of operating systems and mobile applications, Computers, vol. 60, no. 3, pp. 305-320, Mar. 2011.
for estimating the skew angle of document images; a 8. C.-H. Chou, W.-H. Lin, and F. Chang, A binarization method the development of embedded systems faces critical chal- 7. Pei-Han Hsu, Yuan-Hao Chang, Po-Chun Huang, Tei-Wei Kuo,
learning-based binariza-tion method for document im- with learning-built rules for document images produced by lenges in energy consumption and system performance. and David Hung-Chang Du, “A Version-based Strategy for Re-
ages that may have varying degree of brightness; a ma- cameras, Pattern Recognition, vol. 43, no. 4, pp. 1518-1530, We developed new hibernation techniques to shut down liability Enhancement of Flash File Systems,” the ACM/IEEE
chine-learning approach for analyzing the layout structure 2010. the operating system when the battery is running low, and Design Automation Conference (DAC), San Diego, Jun. 5-10,
2011. (Acceptance rate: 22.6%(156/690)) (Top Conference)

of documents; a method for identifying the language type 9. X.-R. Lin, C.-Y. Guo, and F. Chang, Classifying textual com- to resume the operating system e ciently by stealing the 8. Yuan-Hao Chang, Ping-Yi Hsu, Yung-Feng Lu, and Tei-Wei
of textual components, etc. Many of the above methods ponents of bilingual documents with decision-tree support idle time of storage devices. We also developed new tech- Kuo, “A Driver-Layer Caching Policy for Removable Storage
th
have turned into ingredients of an OCR software system. vector machines, The 11 International Conference on Docu- nologies to adjust the CPU frequencies and the devices’ Devices,” ACM Transactions on Storage, vol. 7, no. 1, pp. 1:1-
One exciting application of this system is to transform ment Analysis and Recognition, September 2011. power modes for energy saving according to the system 1:23, Jun. 2011.
hundreds of thousands volumes of Chinese legacy docu- 10. F. Chang and C.-C. Liu, Solving large-scale multi-label SVM runtime information. Now we are investigating the possi- 9. Yuan-Hao Chang, Cheng-Kang Hsieh, Po-Chun Huang, and Pi-
rd
ments into digital contents. problems with a tree decomposition approach, The 3 Asian bility to develop new fast hibernation technologies with Cheng Hsiu, “A Caching-Oriented Management Design for the

Conference on Machine Learning, November 2011. ner hibernation granularity to minimize the hibernation Performance Enhancement of Solid-State Drives,” ACM Trans-
overheads, and exploring the interplay between hardware actions on Storage, vol. 8, no. 1, pp. 3:1-3:21, Feb. 2012.
resources for the developing of advanced resource sched- 10. Po-Chun Huang, Yuan-Hao Chang, and Tei-Wei Kuo, “Joint
uling technologies to minimize the energy consumption Management of RAM and Flash Memory with Access Pattern
of embedded systems. Considerations,” the ACM/IEEE Design Automation Confer-
ence (DAC), San Fracisco, Jun. 3-7, 2012. (Top Conference)
研究人員
70 Research Faculty
71