[Charlie] Yichuan Tang


PhD student in the Machine Learning group at the University of Toronto.

Advisors: Geoffrey Hinton and Ruslan Salakhutdinov.


Master of Mathematics in Computer Science, University of Waterloo, 2010.

Bachelor of Applied Science in Mechatronics Engineering, University of Waterloo, 2008.


Research interests: machine learning, vision, and cognitive science
Contact: tang@cs•toronto•edu





Deep Mixtures of Factor Analysers

Deep Mixtures of Factor Analysers (ICML 2012)
Yichuan Tang, Ruslan Salakhutdinov and Geoffrey Hinton
In 29th International Conference on Machine Learning, Edinburgh, Scotland, 2012.
 [pdf] [poster] [bibtex]



Deep Lambertian Networks

Deep Lambertian Networks (ICML 2012)
Yichuan Tang, Ruslan Salakhutdinov and Geoffrey Hinton
In 29th International Conference on Machine Learning, Edinburgh, Scotland, 2012.
 [pdf] [poster] [bibtex]



Robust Boltzmann Machine for Recognition and Denoising

Robust Boltzmann Machine for Recognition and Denoising (CVPR 2012)
Yichuan Tang, Ruslan Salakhutdinov and Geoffrey Hinton
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012, Providence, Rhode Island
 [pdf] [poster] [bibtex]



Multiresolution Deep Belief Networks

Multiresolution Deep Belief Networks (AISTATS 2012)
Yichuan Tang and Abdel-rahman Mohamed
Proceedings of the 15th International Conference on Artificial Intelligence and Statistics (AISTATS), 2012, La Palma, Canary Islands
[pdf] [poster] [bibtex]



Data Normalization in the Learning of RBMs

RBM learning using approximate ML methods is known to be hard when the input data is not sparse. We present an extremely simple data normalization technique to improve learning. The method consists of only 3 additional lines of code and is akin to training on zero-meaned data.


Data Normalization in the Learning of RBMs
Yichuan Tang and Ilya Sutskever
Technical Report, Department of Computer Science, University of Toronto. 2011
 [pdf] [poster] [bibtex]



Gated Boltzmann Machine for Recognition under Occlusion

Object recognition in the real world must deal with occlusions and clutter. We investigate a modified version of Deep Boltzmann Machine (DBM) called Denoising Gated Boltzmann Machine (DGBM) which uses feedback dynamics to achieve lower recognition error rates.


Gated Boltzmann Machine for Recognition under Occlusion
Yichuan Tang
In NIPS Workshop on Transfer Learning by Learning Rich Generative Models, Whistler, Canada, 2010.
 [pdf] [poster] [bibtex]

DGBM can be used for denoising:
Flash Not Available


Deep Networks for Robust Visual Recognition (ICML 2010)

We show that by performing denoising using a Deep Belief Net prior to classification leads to improved accuracy on noisy and occluded test data.


Deep Networks for Robust Visual Recognition
Yichuan Tang and Chris Eliasmith
In 27th International Conference on Machine Learning, Haifa, Israel, 2010.
 [pdf] [poster] [bibtex]















Patent:

Tang, Y., Zhou, H. "Method and apparatus for identifying regions of different content in an image", U.S. Patent 7,840,071, Nov. 23, 2010.

Chess engine I wrote as a high school CS project: Here, circa 2003.








I completed my Masters of Mathematics degree in 2010 at the University of Waterloo in Computer Science.
My supervisor was Chris Eliasmith.

Prior to my Masters degree, I was part of the class of 2008 in Mechatronics Engineering at the University of Waterloo. I went to the excellent Thomas Worthington HS and Forest Hill Collegiate Institute.
I'm also a Canadian master at the game of chess and two time ('01 & '02) high school chess champion in the state of Ohio.




Last updated, May 2012