Nick Bartlett

Ph.D. Candidate, Statistics, Columbia University

Research Interests

Currently I am focused on methods in Bayesian nonparametrics. Bayesian nonparametrics present a framework for developing flexible, expressive models while preventing overfitting through the use of hierarchy and prior information. Unfortunately, like most nonparametric methods, the models usually require the learner to “remember” all of the data all of the time. Clearly, if we are interested in a learner capable of replicating human knowledge aquisition, the representation of a model must have a memory upper bound. Even if our motives are not to replicate human learning, models where the complexity grows with the number of data points are not practical for data on a massive scale. I am interested in the development of models which use the ideas of Bayesian nonparameterics while requiring a constant space model representation.

My Research Statement