UCSC-SOE-13-09: Bagging During Markov Chain Monte Carlo for Smoother Predictions

Herbert K. H. Lee
07/07/2013 11:30 PM
Applied Mathematics & Statistics
Making good predictions from noisy data is a challenging problem. Methods to improve the robustness of predictions include bagging and Bayesian shrinkage approaches. These methods can be gainfully combined by doing bootstrap resampling during Markov chain Monte Carlo (MCMC). The result is smoother predictions that are less affected by outliers or particularly noisy data points. An example is provided in predicting from a neural network model.