Interactive Machine Learning: Leveraging Human Intelligence

Advances in processor speed and in machine learning algorithms have brought machine learning within the interactive time scale. It is now possible on many problems for a user to annotate data, train a classifier and receive feedback on the classification in 4-5 seconds. This very tight feedback loop on machine learning opens many possibilities for user interfaces that interactively manage far more information than the user can visualize. This not only changes the way we might interact with information but also changes the way we pose our machine learning problems. Having a user in a tight loop changes the distribution of training data as well as imposes constraints on training algorithms. This talk will explore these issues as well as present preliminary data on how humans and machine learning can interact.