# Intro

Hello Guys,

My name is Ga Wu(officially), but I prefer to be called Wuga.

I was a Ph.D. student and research assistant at Oregon State University(United States) and have transferred to University of Toronto(Canada) last fall(2016). I am currently working with Professor Scott Sanner. I got my master degree from Australian National University(Australia) and got my bachelor degree from Northwest Normal University(China).

My research topic is to explore and understand the structure of the deep neural networks. Since deep learning becomes more and more practical, from early years environment perceptions(deterministic) to recent near realistic data(image/sound) generation(variational inference), I want to answer some open questions like “How can we do exact inference through exploiting deep network structures?”.

I am also interested in the entire field of artificial intelligence: Machine Learning, Reinforcement Learning(AGENTS), Deep Learning(Computer vision) and Nature Language Processing. I have lots of experience in those fields due to years of doing projects.

# Publications

G.Wu, S.Sanner, and R.F.S.C.Oliveira Bayesian Model Averaging Naive Bayes: Averaging over an Exponential Number of Feature Models in Linear Time. In Proceedings of the 29th Conference on Artiﬁcial Intelligence(AAAI-15). Austin,USA.

# Github Toolbox

Deconvolution Neural Network in Theano

This is probably the only code you can find online to implement convolution deconvolution neural network under python environment! The another available choice is to use modified version of caffe.

Neural Network Implementation in Python

This is a tutorial code to support student to learn to build fully connected neural network to do digit number classification. The exactly same effective method is to use PCA, which can also be found from my pository.

PCA and LDA

This is one example of using PCA and LDA under python environment. The objective of this implementation is to show the difference of this two methods and when the PCA will be failure.

Delaunay Triangulation

This is one java version of delaunary triangulation, which is the inverse map of Voronoi diagram. I didn’t find the tool in java, so I made this one to help people who may has similar requirement. This code also include Kruskal Minimum Spanning Tree!

Monte Carlo Tree Search on Traffic Control

This code is an implementation of monte carlo tree search on traffic control. It was a course project I took in OSU, which was highly interesting to the Professor Alan Fern. But due to time limitation, we didn’t continue this project.

# Skills

- Deep Learning(GAN, VAC, CNN, RNN) with practical experience on (Keras, Tensorflow and Theano)
- Probabilistic Graphical Model (MCMC, Variational Inference)
- Traditional Machine Learning (SVM, Linear/Logistic Regression, etc)
- Classic Planning(A* and heuristic guided variants)
- AI and Reinforcement Learning (Deep Q-learning, Sarsa, Monte Carlo Tree Search)
- Map-Reduce