On simple learning scenarios, understand and teach step by step how artificial neural networks work.

Draw connections inside the network yourself and observe the consequences on the robot’s movements. Thanks to multiple displays (activity values, thickness of connections, colors for learning, etc.), explain visually the learning equations that automatically optimize these connections.

Activate the camera and the intermediate layers of neurons and see how the simple principles detailed above allow the learning of complex behaviors (obstacle avoidance, pushing a ball, etc.) after scaling up to thousands of neurons and hundreds of thousands of connections. You have in front of you all the calculations made, but they are so numerous that we still miss something, isn’t it fascinating?

Discover – still very visually – other Machine Learning algorithms. In particular, the K-nearest  neighbors algorithm, which is part of several high school curricula, is particularly intuitive and  adapted to introduce supervised learning. Then compare its results to those of neural networks, on the same training database.