A0451
Title: A unifying framework for generalized Bayesian online learning in non-stationary environments
Authors: Gerardo Duran-Martin - University of Oxford (United Kingdom) [presenting]
Leandro Sanchez-Betancourt - University of Oxford (United Kingdom)
Alexander Shestopaloff - Queen Mary University of London (United Kingdom)
Kevin Murphy - Google DeepMind (United States)
Abstract: A unifying framework is proposed for methods that perform probabilistic online learning in non-stationary environments. The framework is called BONE, which stands for generalized (B)ayesian (O)nline learning in (N)on-stationary (E)nvironments. BONE provides a common structure to tackle a variety of problems, including online continual learning, prequential forecasting, and contextual bandits. The framework requires specifying three modeling choices: (i) a model for measurements (e.g., a neural network), (ii) an auxiliary process to model non-stationarity (e.g., the time since the last changepoint), and (iii) a conditional prior over model parameters (e.g., a multivariate Gaussian). The framework also requires two algorithmic choices, which are used to carry out approximate inference under this framework: (i) an algorithm to estimate beliefs (posterior distribution) about the model parameters given the auxiliary variable, and (ii) an algorithm to estimate beliefs about the auxiliary variable. It is shown how the modularity of the framework allows for many existing methods to be reinterpreted as instances of BONE, and it allows the proposal of new methods.
