My goal is to outline a lesson that any teacher can use in the classroom or any person interested in a very high level understanding of how AI works can walk through. This is not meant to be an exact representation of how AI truly works, but simply give intuition as to how it works. I have been a Math, SAT, ACT, ISEE tutor for close to a decade and work in machine learning research.
Pre-requisites: know what a probability is.
There are 2 sub-lessons, 1 smaller one and 1 larger one. All lessons will be under the scope of computer vision problems — object detection.
- Supervised learning vs Unsupervised learning
- Training a machine learning model
Machine learning problems are often broken into two categories, supervised and unsupervised problems. Supervised problems are where you give the model examples of something and then expect it to be able to predict that thing later on an unseen image. Unsupervised problems are where you have a bunch of images and you try to figure out which ones are most closely related (not based on anything except what you can see) and then group them without knowing what the final class you are trying to predict actually is.
I will now show you a series of shapes and a name for the shape.
These shapes above are called zhags.
These shapes above are called flarks.
Now I will present you with an object and you tell me if it’s a zhag or a flark. There is a hidden rule that categorizes zhags and flarks. Your job is to learn that rule.
This is a zhag. If you guessed that, awesome! You learned a successful model.
But maybe now you get an object that doesn’t fit exactly what you thought.
This is a flark.
Little did you know, the hidden rule is if the shape has any curve at all it is a flark. This is why sufficient training data is so important to machine learning problems! If this was a missing training data point in an autonomous vehicle this could cost someone their life.
Say we have a set of images and strictly using the images and no previous knowledge we need to place them on the xy-plane where their distance between each other represents how different they are from one another.
Here are a group of images.
Now we are meant to place these on the xy-plane. Here’s a possible iteration of this.
So if I now said, group these into two sets you probably would do this one of two ways.