Page 159 - My FlipBook
P. 159
Research Fellow 呂及人 研
究
Chi-Jen Lu 人
員
Ph.D., Computer Science, University of Massachusetts at Amherst, United States Faculty
T +886-2-2788-3799 ext. 1820 E cjlu@iis.sinica.edu.tw
F +886-2-2782-4814 W www.iis.sinica.edu.tw/pages/cjlu
・ Assistant Research Fellow, Institute of Information Science, Academia Sinica (1999/8-
2003/10)
・ Associate Research Fellow, Institute of Information Science, Academia Sinica (2003/10-
2008/11)
Research Description
A common practice in machine learning is to learn a model (e.g. a classifier) in a batch way, in which one first collects a set of training
examples and then learns a model from this training set. Afterwards, the learned model is used for all the future testing data, but it remains
xed without being further updated. While this is su cient for many applications considered today, it may not work well for others. In fact,
this does not seem to be the way we humans usually learn. This motivates the study of learning in the online setting, in which the learning
process never stops as long as new data keeps coming. It has evolved into a fruitful research area of machine learning, and my works mainly
cover the following two aspects.
The rst is to identify existing learning algorithms which have been widely used but work in the batch setting, and transform them to work
in the online setting. In this direction, we have successfully developed online versions of several powerful batch algorithms, such as boosting
algorithms, principal components analysis, and tensor decomposition. The second aspect is on the topic known as the online decision
problem, which captures the dilemma we often have to face: to make repeated decisions in unknown and changing environments and su er
the consequences of our decisions. For this problem, we identi ed natural scenarios in which better online algorithms can be designed, and
moreover, we extended the scope of the problem in several directions and develop new algorithms correspondingly. In addition, we also
found new applications of the problem in the area of game theory.
1. Yi-Shan Wu, Po-An Wang, and Chi-Jen Lu, "Lifelong Publications
optimization with low regret," Proceedings of the 22nd
International Conference on Artificial Intelligence and Statistics 7. Chen-Yu Wei, Yi-Te Hong, and Chi-Jen Lu, "Tracking the Best
(AISTATS), April 2019. Expert in Non-stationary Stochastic Environments," Proceedings
of the 30th Annual Conference on Neural Information Processing
2. Jun-Kun Wang, Chi-Jen Lu, and Shou-De Lin, "Online linear Systems (NIPS), December 2016.
optimization with sparsity constraints," Proceedings of the 30th
International Conference on Algorithmic Learning Theory (ALT), 8. Po-An Chen and Chi-Jen Lu, "Generalized mirror descents in
March 2019. congestion games," Artificial Intelligence, volume 241, pages
217-243, December 2016.
3. Chi-Ning Chou, Kai-Min Chung and Chi-Jen Lu, "On the
Algorithmic Power of Spiking Neural Networks," The 10th 9. Chun-Liang Li, Hsuan-Tien Lin, and Chi-Jen Lu, "Rivalry of Two
Innovations in Theoretical Computer Science (ITCS 2019), Families of Algorithms for Memory-Restricted Streaming PCA,"
January 2019. Proceedings of the 19th International Conference on Artificial
Intelligence and Statistics (AISTATS), May 2016.
4. Jen-Hou Chou and Chi-Jen Lu, "The communication complexity
of graphical games on grid graphs," Proceedings of the 14th 10. Yi-Te Hong and Chi-Jen Lu, "Online Learning in Markov
Conference on Web and Internet Economics 14th Conference on Decision Processes with continuous actions," Proceedings of the
Web and Internet Economics (WINE), December 2018. 26th International Conference on Algorithmic Learning Theory
(ALT), Lecture Notes in Artificial Intelligence, October 2015.
5. C h e n - Yu We i , Yi - Te H o n g , a n d C h i - J e n L u , " O n l i n e
Reinforcement Learning in Stochastic Games," Proceedings of Brochure 2020
the 31st Annual Conference on Neural Information Processing
Systems (NIPS), December 2017.
6. Po-An Wang and Chi-Jen Lu, "Tensor Decomposition via
Simultaneous Power Iteration," Proceedings of the 34th
International Conference on Machine Learning (ICML), August
2017.
157
究
Chi-Jen Lu 人
員
Ph.D., Computer Science, University of Massachusetts at Amherst, United States Faculty
T +886-2-2788-3799 ext. 1820 E cjlu@iis.sinica.edu.tw
F +886-2-2782-4814 W www.iis.sinica.edu.tw/pages/cjlu
・ Assistant Research Fellow, Institute of Information Science, Academia Sinica (1999/8-
2003/10)
・ Associate Research Fellow, Institute of Information Science, Academia Sinica (2003/10-
2008/11)
Research Description
A common practice in machine learning is to learn a model (e.g. a classifier) in a batch way, in which one first collects a set of training
examples and then learns a model from this training set. Afterwards, the learned model is used for all the future testing data, but it remains
xed without being further updated. While this is su cient for many applications considered today, it may not work well for others. In fact,
this does not seem to be the way we humans usually learn. This motivates the study of learning in the online setting, in which the learning
process never stops as long as new data keeps coming. It has evolved into a fruitful research area of machine learning, and my works mainly
cover the following two aspects.
The rst is to identify existing learning algorithms which have been widely used but work in the batch setting, and transform them to work
in the online setting. In this direction, we have successfully developed online versions of several powerful batch algorithms, such as boosting
algorithms, principal components analysis, and tensor decomposition. The second aspect is on the topic known as the online decision
problem, which captures the dilemma we often have to face: to make repeated decisions in unknown and changing environments and su er
the consequences of our decisions. For this problem, we identi ed natural scenarios in which better online algorithms can be designed, and
moreover, we extended the scope of the problem in several directions and develop new algorithms correspondingly. In addition, we also
found new applications of the problem in the area of game theory.
1. Yi-Shan Wu, Po-An Wang, and Chi-Jen Lu, "Lifelong Publications
optimization with low regret," Proceedings of the 22nd
International Conference on Artificial Intelligence and Statistics 7. Chen-Yu Wei, Yi-Te Hong, and Chi-Jen Lu, "Tracking the Best
(AISTATS), April 2019. Expert in Non-stationary Stochastic Environments," Proceedings
of the 30th Annual Conference on Neural Information Processing
2. Jun-Kun Wang, Chi-Jen Lu, and Shou-De Lin, "Online linear Systems (NIPS), December 2016.
optimization with sparsity constraints," Proceedings of the 30th
International Conference on Algorithmic Learning Theory (ALT), 8. Po-An Chen and Chi-Jen Lu, "Generalized mirror descents in
March 2019. congestion games," Artificial Intelligence, volume 241, pages
217-243, December 2016.
3. Chi-Ning Chou, Kai-Min Chung and Chi-Jen Lu, "On the
Algorithmic Power of Spiking Neural Networks," The 10th 9. Chun-Liang Li, Hsuan-Tien Lin, and Chi-Jen Lu, "Rivalry of Two
Innovations in Theoretical Computer Science (ITCS 2019), Families of Algorithms for Memory-Restricted Streaming PCA,"
January 2019. Proceedings of the 19th International Conference on Artificial
Intelligence and Statistics (AISTATS), May 2016.
4. Jen-Hou Chou and Chi-Jen Lu, "The communication complexity
of graphical games on grid graphs," Proceedings of the 14th 10. Yi-Te Hong and Chi-Jen Lu, "Online Learning in Markov
Conference on Web and Internet Economics 14th Conference on Decision Processes with continuous actions," Proceedings of the
Web and Internet Economics (WINE), December 2018. 26th International Conference on Algorithmic Learning Theory
(ALT), Lecture Notes in Artificial Intelligence, October 2015.
5. C h e n - Yu We i , Yi - Te H o n g , a n d C h i - J e n L u , " O n l i n e
Reinforcement Learning in Stochastic Games," Proceedings of Brochure 2020
the 31st Annual Conference on Neural Information Processing
Systems (NIPS), December 2017.
6. Po-An Wang and Chi-Jen Lu, "Tensor Decomposition via
Simultaneous Power Iteration," Proceedings of the 34th
International Conference on Machine Learning (ICML), August
2017.
157
