MIT-xPRO-DSx-DECISION-BOUNDARY-AND-DEEP-NETWORK. Case Study 3.2: Alpha Go Instructor: Ankur Moitra Activity Type: Optional Case Study Description: Playing with one or two-layer perceptrons to get a feeling of their decision boundaries. Decision boundary of a deep neural network Why this Case Study? The case study shows the advantage of adding nonlinearity and using deeper neural networks over shallow ones. Self-Help Documentation In this document, the following is discussed. Layering perceptrons are mysterious. What types of patterns can deep neural networks recognize? And in what ways are they more powerful than just a single perceptron? In this case study, we will explore these questions in two dimensions, so that we can visualize them easily. What happens in higher dimensions is much more complex, and an area of active research. Download Self-Help Documentation Time Required: The time required to do this activity varies depending on your experience in the required programming background. We suggest planning somewhere between 1 & 3 hours. Remember, this is an optional activity for participants looking for hands-on experience.