Machine learning and AI are no longer far-future innovations — they’re the technologies revolutionizing our work and lives right now. For engineers to tackle real-world problems, they need to go beyond reading cutting-edge research and learning about the latest algorithms. It’s time for a practical approach that brings AI capabilities and impact together.

If you’re interested in using machine learning to enhance your product, it’s important to look at how the entire development process works. Understand what happens before, during and after training your models — and how each step can lead to success or failure.

The ads machine learning and AI team at Meta developed a video series to help engineers and researchers apply their machine learning skills to real-world problems. The series breaks down the machine learning process into six steps:

1. Problem definition
2. Model
3. Evaluation
4. Features
5. Model
6. Experimentation

Following each step, you’ll learn how to successfully apply machine learning to your product or use case — including unexpected lessons that are important in an applied setting. You’ll gain new knowledge of the machine learning process, the importance of making the right decisions at each step, and how using machine learning models can help deliver business outcomes.

Happy learning!

In this first lesson, explore best practices for defining the problem. The right setup is often more important than the choice of algorithm. A few hours spent at this stage in the process can save several weeks of work downstream, preventing you from solving the wrong problem.

Preparing the training data is a core part of a machine learning engineer’s job. It’s an active, not passive, part of machine learning research — one of the most valuable variables to create high-quality machine learning systems.

Create a clear plan to evaluate the performance of your model before you start developing more features and iterating on model architectures. This lesson covers how to evaluate your approach.

Time to focus on features. Here, see examples of categorical, continuous and derived features. You’ll learn how to choose the right feature for the right model, and what to look out for — such as changing features, feature breakage, leakage and coverage.

Your next job is to choose the right model for your data and find the algorithm to implement and train that model. This lesson offers tips such as:

In the last lesson, discover all you need to know about experimentation, or making your experiments actionable. Dive deep into the difference between offline and online experimentation.

With these real-world best practices, you’re ready to apply machine learning to make a difference in the AI community and beyond. Explore opportunities for you to help redefine the boundaries of what’s possible with a job in machine learning.

This post, originally published on May 7, 2018, was updated on September 20, 2024.