Page 7 - profile2014.indd

所長的話

We lead a major research effort in Chinese language pro- Review, ACM TechNews, Phys.org, and Popsci.com. Our
cessing. The IIS developed the first balanced Sinica Cor- colleagues and Macronix Co., Ltd. have proposed a heal-
pus with more than 5 million words and part-of-speech leveling technique that can significantly enhance the
tagging. Licenses have been issued to major research lifetime of next-generation 3D flash-memory chips; the
groups and companies around the world, including Ya- description was nominated as the best paper candidate
hoo and Apple. Our intelligent Chinese input system, of the ACM/IEEE Design Automation Conference 2014.
“GOING”, has been used by more than a million peo-
ple in Taiwan. We have also developed a Chinese word Our research on bioinformatics has also flourished. As
segment system that has processed more than a million most cancer biomarkers are glycoproteins and mem-
documents. In addition, our Chinese question answer- brane proteins, we are leading the world in developing
ing system has won at major international competitions. an automated glycoprotein identification system that is
Currently, the group is developing sophisticated Chi- capable of analyzing large-scale datasets. Furthermore,
nese natural language tools combining pattern match- we have developed a variety of prediction methods for
ing and machine learning techniques to provide more membrane proteins. We played an early and pivotal role
concise and accurate results to users. in analyzing chromosome 4 in the international Human
Proteome Project. In the field of metabolomics, we have
Over the past few years, more than 30 principal inves- succes
tigators at the IIS have racked up a number of impres-
sive accomplishments across various research fields in
computer science. In 2010, our colleagues developed a
blurred license plate image recognition system, which is
the first system in the world that can solve the associat-
ed problems in a systematic manner. The technique was
transferred to the Industrial Technology Research Insti-
tute in the same year. In addition, the same group’s
efficient multimedia technique for video inpainting to
fix damaged and old films was made available (with the
relevant software) to the National Archives Administra-
tion via technology transfer in 2011. Regarding embed-
ded database systems, we have proposed multi-version
index techniques, which can significantly improve the
data collecting capability of sensors deployed in cyber-
physical systems. In 2012, the virtualization team devel-
oped HQEMU, a high-performance and re-targetable
dynamic binary translation system, which exhibits high
performance and is the first reported case of successful-
ly combining QEMU and LLVM. In 2013, our colleagues
developed an influence maximization algorithm to find
optimal solutions in online social networks, which was
published in ACM KDD, and featured by MIT Technology

7