Following along on the machine learning education path, you may have seen some of these concepts before, but as a quick refresher or for those of you just joining us, first, we need to establish a framework between the relationships of artificial intelligence, machine learning, and deep learning.

In reality, these are all tightly related, but at the broadest definition, everything here is artificial intelligence or AI. Simply put, this is the ability of computers to learn and mimic human behavior. A classic example of this is a chess bot, in which a computer is able to provide a human like experience against another person playing the game of chess.

But at this level, the machine side of the equation is simply governed by rules engines or other simple algorithms. And isn't able to truly grow on its own over time. In the case of chess, it's given a set of moves and told to iterate through all possible scenarios to find the best one for itself.

The next step in the onion that is computer learning, or the next step down the rabbit hole if you were, is machine learning. This is where a computer can learn without explicit programming. In other words, if a computer is given a model and told roughly what a well-formed model looks like, it will draw its own inferences and conclusions and develop the model until it's a pretty accurate way of how to make decisions on its own.

For example, instead of telling the computer how to make chess moves, the chess bot would simply observe a couple of grandmasters play for a few rounds. The computer would be told who was winning at every stage, and it would attempt to draw its own rules on how to play chess. And over time, through machine learning, it will become a fantastic chess bot simply through inferring the rules and being told what a good move is through a health check of, "This player is winning." And finally at the core of all machine based intelligence is the field known as deep learning.

Although this is a subset of machine learning which is internal subset of artificial intelligence, this is a very important subset as this can be viewed as one of the end states or the final goals of all computer-based intelligence. This is where the machine attempts to discover its own systems, its own health, its own relations.

The classic example of this is YouTube's categorization bot. In earlier years, Google set a bot loose on YouTube and said, "Hey, make some categories." And this bot, without knowing anything at all, discovered cat videos simply because a similar sized creature is doing similar things, AKA looking cute, and categorized them all together. And made a cat video category without even knowing what a cat is.

For example, at level one, you have a programmer who is able to leverage out of the box materials. This is someone who is able to integrate an API, such as Google's Vision API or a more text-based service, such as Amazon's Comprehend API. This level in particular was covered in an earlier class in this learning path.

So if you have an interest there, feel free to check it out. I personally have also narrated that. So I highly recommend it, of course, but levels two and three are where we will focus this course. Levels two and three are varying amounts of making your own model and training it. Level two more focuses on pre-made frameworks, such as available in Google's AutoML features. And level three is more such as using Amazon SageMaker, where you're in a Python or other code notebook and programming the models yourself.

Level four is beyond the scope of this class. This is where you're making machine learning models from a more fundamental level. Maybe you're working with a piece of hardware such as a TensorFlow processing unit, but we won't be covering that.

For this class, we'll mostly be focusing on how to make your own models and how to train and pick the right type of model for a level two or three user. To dive into the meat of it, let's talk about what a machine learning model actually is.

So as you may already know, a machine learning model is a computer algorithm that automatically improves itself over time through experience and exposure to different situations. These algorithms build mathematical models off sample data, sometimes called training data. And after enough training, they're able to make predictions or inferences on their own.

So, in other words, after you train them, you might not use an explicit rule set. They're still able to make highly accurate predictions based on their observation or training data.

Lectures

Course Introduction - Models - Understanding Training Data Sets - How to Choose? - Case Study: Home Prices - Case Study: Heart Disease - Case Study: Animal Classification - Case Study: Targeted Marketing