Intro to Training, Evaluating & Fine-tuning an AI

You will learn how to:

• Understand what supervised learning is and how machines learn from labeled examples.
• Build intuition for linear models and how they make predictions from input features.
• Represent data in ways that models can use, including feature vectors and simple preprocessing.
• Train your first machine‑learning model and see how changing parameters affects performance.
• Evaluate models using metrics like accuracy, loss, and train/validation splits.
• Recognize overfitting and underfitting, and learn how to avoid them.
• Improve a model through fine-tuning, adjusting parameters, and iterating on design choices.
• Understand the full workflow of building an AI system—from data to training to evaluation.
• Develop the confidence to read model outputs, diagnose issues, and refine your approach.

You should already be comfortable with:

• Prior coding experience from Joy of Coding: Intro to Data Visualization and AI or its equivalent
• Algebra 1 and Geometry
  • Algebra 2 is helpful (especially vectors and matrices) but not required
• Spending 4–6 hours per week working steadily and submitting your best work
• Asking the instructional team for help when you’re stuck or want to go deeper

The course is online and asynchronous.

A personal computer and access to the internet will be required for each lesson.

By completing the coding assignments and the accompanying reflection activities, you’ll earn an online certificate from the Department of Electrical and Computer Engineering at the University of Michigan.

For this course, you can earn online certificates in:

• Introduction to Linear Supervised Learning
• Introduction to Training, Evaluating and Finetuning an AI

These not‑for‑credit certificates of accomplishment let you show colleges the skills you’ve learned and your ability to learn in a self-regulated manner.

“Before this course, I had no idea how machines actually ‘learn.’ Training my first supervised model made the whole idea of labeled data finally make sense.”

“Seeing how a tiny change in parameters could improve or break a model was eye‑opening. It felt like I was really steering the learning process.”

“I used to think AI was a black box. Now I can read model outputs, spot overfitting, and explain what’s going wrong—and how to fix it.”

“Building a model from raw data all the way to evaluation made me feel like I understood the full AI workflow, not just the buzzwords.”

“Finetuning my model was the moment everything clicked. I realized I could actually improve an AI system on my own, not just run someone else’s code.”