Convolutional Neural Networks (CNNs)

Nora Shipp & Nick Mondrik

Why CNNs?

Problem:

How do we classify an image with neural networks?

Solution:

Use convolutions to find spatial features useful for classification!

How they work

Use different kernels to identify different features

How they work

Example: Number Recognition

http://scs.ryerson.ca/~aharley/vis/conv/'

Summary

1. Convolutions enhance spatial features in images
   
   
2. Nodes in CNN learn kernels useful in classifying images
   
   
3. Astronomical example: Galaxy/Star classification