Welcome to Machine Learning.

If you have been paying attention, you have probably heard about Machine Learning or Artificial Intelligence as some of the most exciting new technology around. But what is it, and how does it differ from traditional programming?

A person telling a robot to tie it's shoes.

Most programming you've done requires you to specify exactly how to complete a task line by line. Machine learning refers to programming techniques that use statistics to allow the program to "learn" and improve at its task.

This is accomplished by building a model with several adjustable parameters that can attempt to accomplish a task. The model learns by testing repeatedly on the task, then optimizing the parameters toward the expected result. Over time, the approximation made by the model will get closer and closer to the actual expected value.

Machine learning problems can be broken down into a few categories. The main one we'll cover in this course is supervised learning. Supervised learning refers to tasks that have a metric for success, whether it's labeled data or rewards for performing well at the task.

On the other side, unsupervised learning refers to problems where the model has no guidance and must infer patterns and features in the data independently.

How Does a Model "Learn"?

The statistical techniques to perform machine learning tasks have existed for decades, but only recently has the computing power necessary to make it practical become available.

Data is fed into the model. The parameters inside the model take that data and return an answer.
The model's guess is then compared with the expected outcome. (Was the image a cat or a horse? Did the ghost eat Ms. Pac Man?)
The parameters are optimized in the direction of the expected outcome.

This process is repeated with new data until the model has optimized its parameters to be as likely as possible to guess the right answer.

What is a Machine Learning Model?

The Model in a Machine Learning program is the combination of mathematical functions and parameters that the program updates as it learns. This will often include a Neural Network - a series of connected functions and parameters that are well suited to adapting to various tasks.

Models are built and trained for different tasks, so the data that each model reads and the output that they produce will vary based on the model's purpose. Teachable Machine's models are primarily used for Classification problems - deciding which of a number of predetermined classes an input fits into. You will be creating a model that will classify whether the current webcam input is a thumbs-up (Yes) or a thumbs-down (No).

The Jupyter Notebook that you will be setting up and using to do this project will walk you through the process of collecting data, training a network, and then testing the model that you trained. Move into the next lessons to learn about how to make this project.