CSAIL Event Calendar: Previous Series
Fast Online Active Learning
The online active learning problem is a realistic scenario, begging several interesting computational questions. Online or sequential learning is a framework in which training examples are received one at a time, and the learner must make a prediction at each time-step. The learner cannot store all previously seen examples and then apply a batch learning algorithm to them, but must instead intelligently summarize its observations. Additionally, the time complexity of the belief update step should be constrained against scaling with the number of past examples, in order for the algorithm to be effective in the online setting. Active learning is a situation in which the labels of the examples are missing, yet available at some cost. Beyond just minimizing the number of examples needed to learn a concept, the goal is to minimize the number of labels that the algorithm needs to check, in doing so. In fact, intelligent choices of which examples to label can even reduce the total number of examples needed for learning.