The field of “Artificial Intelligence” has a long history of over-promising and under-delivering. Each cycle of the marketing hype tweaks the name a little while continuing to over-promise. We’re a long way from Artificial General Intelligence where an intelligent agent can understand and learn anything a human can. However there have been some good advances in recent years applying AI to narrower fields.
Where conventional software is programming by defining steps, Machine Learning is programming by example. You give the computer a load of labelled data and it builds a statistical model of the training data in order to make predictions (or inferences) on new data. There are a bunch of different approaches to building these models and I’m going to get started with Deep Learning which uses neural networks.
FastAI is a very helpful library for getting up and running quickly with Deep Learning. They have a free online course which I’ve been doing. Let’s try out some basic Deep Learning using the fast.ai libraries in a Jupyter Notebook. I ran mine on Paperspace using the “Paperspace + Fast.AI” runtime template.
Let’s build something that CATegorises some big cats (pun intended). It’ll differentiate between Bengali tigers, White tigers and cute tiger teddys.