Supervised learning trains on labeled examples โ a model learns to predict the right answer by studying many question-and-answer pairs. Every question you ask and truthful answer you get is exactly that: a labeled clue narrowing down the correct output.
Unsupervised learning finds structure in data with no labels and no right answer โ nobody tells the algorithm what the groups should be. It just looks at similarities and figures out its own categories, exactly like you're about to do with these faces.
Tap a face to cycle it through your groups. Group however you like โ hair, accessories, skin tone, anything.
Reinforcement learning is trial-and-error with rewards โ an agent tries an action, and success or failure changes what it's allowed to try next. Your growing block limit is the reward signal: the agent (your teammate on the mat) earns more freedom to act the better it performs.
There's more than one route to the goal. Place a block, then press Run.
Blocks you can queue this round: 1
Each clean run lets you queue one more block next round โ crash and it resets to 1. The clock starts on your first block โ 90 seconds to reach the goal.