What Kids Learned at Our Spring Camp-AI trainers (Day 4)

Today, students explored how to train their own AI models to recognize patterns in images.

They built their own models, tested them, and improved them through multiple iterations. Students would train a model, observe how it performed, and then investigate why it did or did not work as expected. Based on their observations, they adjusted their training data and tried again. In many cases, it took several attempts to achieve reliable results.

Through this process, students developed a deeper understanding of what makes an AI model successful. They explored factors such as the amount of training data, the importance of including variations, and how background elements can sometimes confuse a model.

Students also noticed that some models are more challenging to train than others. For example, distinguishing between colors can be difficult due to overlap and variation, which may lead to inconsistent results. We discussed how defining a range can help improve accuracy. In contrast, models like thumbs up vs. thumbs down were easier to train because the differences are more distinct and consistent.

Students then integrated their image recognition models into block-based coding projects. During this process, they learned the importance of including a “none” condition so the program can respond appropriately when an input does not match any trained category.

We also spent time in Minecraft, where students explored patterns in a different context. As they built and interacted within the game, we connected these experiences to how systems recognize and respond to patterns in real-world environments, reinforcing the idea that pattern recognition extends beyond AI models.