Instance-Based Learning

Also known as Memory Based Algorithm. It is a family of learning algorithms that do not perform explicit generalization, rather, the algorithm compares new problem instances with instances which were seen in training and have been stored in memory. It is called that because the hypothesis created by the algorithm is created directly from the training data instances themselves. This means that the hypothesis complexity grows as the size of the data grows. Example of an instance-based algorithm is KNN (k-nearest-neighbors). Illustration of KNN - the red triangles and the blue squares are previously seen instances, the algorithm should classify the green circle with a question mark based on the previous instances.