How Machine Learning Works

Introduction to Machine Learning (and Review of AI)

Welcome back! Yesterday, we experimented with artificial intelligence (AI), which was a set of techniques our computers use to act on their own, making decisions much like us humans would. We saw examples of classification and prediction which all went a step further into AI and used machine learning.

Machine learning is a way in which computers learn 👨🏾‍🏫 rules on their own using historical data to train, practice, and then make predictions. Today, we’ll explore further just how computers learn. We’ll relate the process to human anatomy, our brain 🧠 and neurons, and we’ll talk about how we can help computers practice and learn by providing strong 💪 and diverse 🌎 datasets!


Machine Learning is Different

Unlike traditional programming where we provide all of the explicit rules to the computer, machine learning is rooted in the idea that computers can learn the rules for themselves (with a few hints from us, of course). This way, we don’t have to think up every different possible input. Our trained model can make best guesses on its own, using rules it created!


Models and Training

A model is at the heart of any machine learning application. It’s the framework the system use for making inferences (or predictions). Because our machine learning model is used for artificial intelligence (used to mimic a human), it’s structure is a lot like our human anatomy. In fact, like our brain 🧠, we use neurons and neural networks in our machine learning models.

To make machine learning work, our models need to be trained. We give them “practice problems” and then let the models guess and check, learning and improving along the way (very similar to the process you might use to study for a test)!


Transforming our Inputs

There are only certain ways our model can accept inputs! It’s on us to make sure the data inputs are structured in a friendly format for our model. We flatten and apply filters to our inputs to help pick out the important features that our model might want to see 👀 when making a prediction.


Data, Data, Data!

Our machine learning application and the model at the heart of our system is only as good as its training data, similar to the way you can speak one language, but probably can’t understand too many others! Luckily, we can get data from lots of sensors in the world! As more and more of our devices become smart and connected, there’s more data to collect! But we have to make sure that the data is clean, well-managed, labeled correctly, and from a diverse background (that means lots of different inputs from lots of different people). Otherwise, our systems could have bias, which isn’t fair to some people :(

We also have to watch out for overtraining where our model only becomes good at identifying the inputs it’s already seen and nothing else!

This is a lot to think about, and it's one of the biggests challenges in ML today!


The Machine Learning Workflow

Let’s review everything we need to know for building a machine learning application, from start to finish!


Agenda for the Afternoon