Simple Visualization: How Machine Learns

Original article was published by Kr. Wan on Deep Learning on Medium


Simple Visualization: How Machine Learns

A simple Visualization

Machine Learning is interesting yet boring. The training process is a mystery and sometimes we simply want a direct understanding on what happens during training. The following video is edited by me to show the learning process of four algorithms to classify cats, dogs, and pandas. Though they are very simple, I think it provides a good enough intuition on what happens during training.

Besides the top right algorithm which collapsed after 200 episodes, the rest three all did a decent job on classify and recognize Panda. However, it is apparent that it is harder for those three algorithms to distinguish dogs and cats. What could the problem be? My suspect is the the learning rate is too high, but there could definitely be other reasons. Nevertheless, a visualization does provide a more direct understanding on what happens during training (and it is fun to watch!).