MASSACHUSETTS INST OF TECH CAMBRIDGE CAMBRIDGE United States
In many scenarios, we want our model of the world to be able to grow in complexity as we collect more information from the world around us. This growth reflects that we learn more about the world as we acquire more data. And we wish to explicitly model both rare events as well as the potential for new events or latent outcomes that we have not yet experienced or collected data on. In this project, we have developed new model representations that enable fast and efficient inference, as well as provided and proved error bounds for certain classes of approximation. Our experiments below were on simulated data. We started preliminary work on the Innovian Time-Series Anesthesia Dataset but were given access to that data a few weeks before the project concluded and were not able to run our full experiments in that time. We did not create any new data sets as part of this project.