Deep Learning

 

(TBA)

 

This is an ongoing project and will be updated as soon as I can.

 

One of the most exciting developments in computer science/machine learning recently has been the new algorithms that made deep learning possible. I have started to explore the potential of deep learning in astronomy, such as in galaxy morphological classification and the large-scale structure of the Universe. The note below is a summary of the standard backward error propagation (backpropagation in short) learning algorithm in vanilla artificial neural network (ANN, left) and convolutional neural network (CNN, middle and right). I will put it down in latex soon.

I have been developing my own code:

 

https://github.com/guangtunbenzhu/BGT-Cosmology/tree/master/DeepLearning

 

The vanilla ANN code (ann.py) is fast and robust. The convolutional CNN code (cnn.py) is still work in progress and will be updated soon.

 

This is a nice tutorial on neural network for beginners, 

http://neuralnetworksanddeeplearning.com/

 

There are also a few well-written packages in deep learning, such as

 

 

They have GPU options to speed up the convolution operations drastically.

 

 

Johns Hopkins University

© 2016 Guangtun Ben Zhu

NASA Hubble Fellowship Program