Think of an example of each of the three types of learning (supervised, unsupervised, reinforcement) that babies and young children do. Is learning to play chess supervised, unsupervised, or reinforcement learning? Or could it be more than just one? Recall, from the lecture, Ilya Sutskever's network that can write some English. Was that a supervised learning task, an unsupervised learning task, or a reinforcement learning task? Unsupervised learning tries to create a compact, low-dimensional representation of the input. What might be such a representation when the input is an audio recording of somebody telling a story? The perceptron learning algorithm can only find a good weights vector if a good weights vector exists (obviously). "Good" means that with that weights vector, all training cases of the training set are classified correctly. Can you think of a simple training set (a collection of training cases) for which no good weights vector exists? What would the perceptron learning algorithm do if you gave it that dataset? 2. Now answer the same question, with a training set where every training case has different input values. Make a training set with only one input dimension.