Page 176 - My FlipBook
P. 176
研
人 Research Fellow
員 黃文良 Wen-Liang Hwang
Faculty Ph.D., Computer Science, New York Univeristy, United States
T +886-2-2788-3799 ext. 1609 E whwang@iis.sinica.edu.tw
F +886-2-2782-4814 W www.iis.sinica.edu.tw/pages/whwang
・ Research Fellow, Institute of Information Science, Academia Sinica, Taiwan
(2005/1-present)
・ Associate Research Fellow, Institute of Information Science, Academia Sinica, Taiwan
(1999/1-2005/1)
・ Assistant Research Fellow, Institute of Information Science, Academia Sinica, Taiwan
(1995/1-1999/1)
・ Associate Post Doctoral Researcher, Department of Mathematics, University of
California, Irvine, United States (1993/1-1994/1)
・ Research Assistant, Department of Computer Science, New York University, United
States (1990/1-1993/1)
・ Ph.D., Computer Science, New York University, United States (1989/1-1993/6)
・ M.S., Electrical Engineering, Polytechnic University, United States (1986/9-1989/1)
・ B.S., Nuclear Engineering, National Tsing Hua University, Taiwan (1977/9-1981/7)
Research Description
My research interests are deep neural network analysis and optimization. In deep neural network analysis, I invent the un-rectifying
technique to derive the explicit representation of a network with point-wise piecewise linear activation functions. By now, I focus on using
the technique to derive solutions to problems relevant to learning deep feed-forward networks, invertible deep networks, and networks
containing loops. The inverse problems can be treated by the regularization or by the forward inference through a non-linear network. To
study the invertible deep networks is motivated from observing inverse problem can also be solved by backward inference if the problem
can be represented as an invertible network. In biology, the connectivity of neurons comprises of loops. To discover what information is
encoded along the time via loops motivates my study of the dynamic behaviors of loops. I am also attracted by optimization methods. In
research, I focus on realizing the idea of introducing auxiliary operators (functions) to help solving optimization problems. So far, I do not fully
understand the feasibility of the idea rather than exercising it in cases by cases basis. In one case, I demonstrate operators can be constructed
and applied to separation of a super-position of di ering types of signals from observations of one channel or multiple channels. In my
previous researches, I study instantaneous frequency and singularity detection and characterization methods via wavelets. My representative
shows that both can be simultaneously detected and characterized using complex-valued wavelets. My representative in video coding is the
formulation of the scalable coding as a multi-objective optimization problem and identi es some interesting Pareto points for performance
measurements. This result provides a way to understand the intrinsic di culty of performance evaluation of scaling coding and may inspire
new performance evaluation methods.
Publications
1. Wen-Liang Hwang, Andreas Heinecke, "Un-rectifying Non-linear 6. Xiyuan Hu, Silong Peng, and Wen-Liang Hwang, "EMD
Networks for Signal Representation," IEEE Transactions on Revisited: A New Understanding of the Envelope and Resolving
Signal Processing, volume 68, pages 196-210, December 2019. the Mode-Mixing Problem in AM-FM Signals," IEEE
Transactions on Signal Processing, volume 60, number 3, pages
2. Wen-Liang Hwang, Chia-Chen Lee, and Guan-Ju Peng, "Multi- 1075-1086, March 2012.
Objective Optimization and Characterization of Pareto Points for
Scalable Coding," IEEE Transactions on Circuits and Systems for 7. Silong Peng and Wen-Liang Hwang, "Null Space Pursuit: An
Video Technology, volume 29, number 7, pages 2096 - 2111, July Operator-based Approach to Adaptive Signal Separation," IEEE
2019. Transactions on Signal Processing, volume 58, pages 2475-2483,
May 2010.
3. Wen-Liang Hwang, Ping-Tzan Huang, Bo-Chen Kung, Jinn Ho, and
Tai-Lang Jong, "Frame-based Sparse Analysis and Synthesis 8. Chun-Liang Tu, Wen-Liang Hwang, and Jinn Ho, "Analysis of
Signal Representations and Parseval K-SVD," IEEE Transactions Singularities from Modulous Maxima of Complex Wavelets,"
on Signal Processing, volume 67, number 12, pages 3330 - 3343, IEEE Transactions on Information Theory, volume 51, number 3,
June 2019. pages 1049-1062, March 2005.
4. Wen-Liang Hwang, Keng-Shih Lu, and Jinn Ho, "Constrained 9. Wen-Liang Hwang and Stephane Mallat, "Singularity Detection
Null Space Component Analysis for Semi-Blind Source and Processing with Wavelets," IEEE Transactions on Information
Separation Problem," IEEE Transactions on Neural Networks and Theory, volume 32, number 2, March 1992.
Learning Systems, volume 29, pages 377-391, November 2016.
10. Rene Carmona, Wen-Liang Hwang, and Bruno Torresani,
5. Jinn Ho, Wen-Liang Hwang, "Wavelet Bayesian Network Image "Practical Time-Frequency Analysis," Academic Press, 1998.
Denoising," IEEE Transactions on Image Processing, volume 22,
number 4, pages 1277 - 1290, April 2013.
174
人 Research Fellow
員 黃文良 Wen-Liang Hwang
Faculty Ph.D., Computer Science, New York Univeristy, United States
T +886-2-2788-3799 ext. 1609 E whwang@iis.sinica.edu.tw
F +886-2-2782-4814 W www.iis.sinica.edu.tw/pages/whwang
・ Research Fellow, Institute of Information Science, Academia Sinica, Taiwan
(2005/1-present)
・ Associate Research Fellow, Institute of Information Science, Academia Sinica, Taiwan
(1999/1-2005/1)
・ Assistant Research Fellow, Institute of Information Science, Academia Sinica, Taiwan
(1995/1-1999/1)
・ Associate Post Doctoral Researcher, Department of Mathematics, University of
California, Irvine, United States (1993/1-1994/1)
・ Research Assistant, Department of Computer Science, New York University, United
States (1990/1-1993/1)
・ Ph.D., Computer Science, New York University, United States (1989/1-1993/6)
・ M.S., Electrical Engineering, Polytechnic University, United States (1986/9-1989/1)
・ B.S., Nuclear Engineering, National Tsing Hua University, Taiwan (1977/9-1981/7)
Research Description
My research interests are deep neural network analysis and optimization. In deep neural network analysis, I invent the un-rectifying
technique to derive the explicit representation of a network with point-wise piecewise linear activation functions. By now, I focus on using
the technique to derive solutions to problems relevant to learning deep feed-forward networks, invertible deep networks, and networks
containing loops. The inverse problems can be treated by the regularization or by the forward inference through a non-linear network. To
study the invertible deep networks is motivated from observing inverse problem can also be solved by backward inference if the problem
can be represented as an invertible network. In biology, the connectivity of neurons comprises of loops. To discover what information is
encoded along the time via loops motivates my study of the dynamic behaviors of loops. I am also attracted by optimization methods. In
research, I focus on realizing the idea of introducing auxiliary operators (functions) to help solving optimization problems. So far, I do not fully
understand the feasibility of the idea rather than exercising it in cases by cases basis. In one case, I demonstrate operators can be constructed
and applied to separation of a super-position of di ering types of signals from observations of one channel or multiple channels. In my
previous researches, I study instantaneous frequency and singularity detection and characterization methods via wavelets. My representative
shows that both can be simultaneously detected and characterized using complex-valued wavelets. My representative in video coding is the
formulation of the scalable coding as a multi-objective optimization problem and identi es some interesting Pareto points for performance
measurements. This result provides a way to understand the intrinsic di culty of performance evaluation of scaling coding and may inspire
new performance evaluation methods.
Publications
1. Wen-Liang Hwang, Andreas Heinecke, "Un-rectifying Non-linear 6. Xiyuan Hu, Silong Peng, and Wen-Liang Hwang, "EMD
Networks for Signal Representation," IEEE Transactions on Revisited: A New Understanding of the Envelope and Resolving
Signal Processing, volume 68, pages 196-210, December 2019. the Mode-Mixing Problem in AM-FM Signals," IEEE
Transactions on Signal Processing, volume 60, number 3, pages
2. Wen-Liang Hwang, Chia-Chen Lee, and Guan-Ju Peng, "Multi- 1075-1086, March 2012.
Objective Optimization and Characterization of Pareto Points for
Scalable Coding," IEEE Transactions on Circuits and Systems for 7. Silong Peng and Wen-Liang Hwang, "Null Space Pursuit: An
Video Technology, volume 29, number 7, pages 2096 - 2111, July Operator-based Approach to Adaptive Signal Separation," IEEE
2019. Transactions on Signal Processing, volume 58, pages 2475-2483,
May 2010.
3. Wen-Liang Hwang, Ping-Tzan Huang, Bo-Chen Kung, Jinn Ho, and
Tai-Lang Jong, "Frame-based Sparse Analysis and Synthesis 8. Chun-Liang Tu, Wen-Liang Hwang, and Jinn Ho, "Analysis of
Signal Representations and Parseval K-SVD," IEEE Transactions Singularities from Modulous Maxima of Complex Wavelets,"
on Signal Processing, volume 67, number 12, pages 3330 - 3343, IEEE Transactions on Information Theory, volume 51, number 3,
June 2019. pages 1049-1062, March 2005.
4. Wen-Liang Hwang, Keng-Shih Lu, and Jinn Ho, "Constrained 9. Wen-Liang Hwang and Stephane Mallat, "Singularity Detection
Null Space Component Analysis for Semi-Blind Source and Processing with Wavelets," IEEE Transactions on Information
Separation Problem," IEEE Transactions on Neural Networks and Theory, volume 32, number 2, March 1992.
Learning Systems, volume 29, pages 377-391, November 2016.
10. Rene Carmona, Wen-Liang Hwang, and Bruno Torresani,
5. Jinn Ho, Wen-Liang Hwang, "Wavelet Bayesian Network Image "Practical Time-Frequency Analysis," Academic Press, 1998.
Denoising," IEEE Transactions on Image Processing, volume 22,
number 4, pages 1277 - 1290, April 2013.
174
