Statistical analysis with missing data
Statistical analysis with missing data
Introduction to Reinforcement Learning
Introduction to Reinforcement Learning
On Bayesian model and variable selection using MCMC
Statistics and Computing
Online Model Selection Based on the Variational Bayes
Neural Computation
Suppressing model overfitting in mining concept-drifting data streams
Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining
Model-driven data acquisition in sensor networks
VLDB '04 Proceedings of the Thirtieth international conference on Very large data bases - Volume 30
Multi-armed bandit algorithms and empirical evaluation
ECML'05 Proceedings of the 16th European conference on Machine Learning
Online optimization for variable selection in data streams
Proceedings of the 2008 conference on ECAI 2008: 18th European Conference on Artificial Intelligence
Hi-index | 0.00 |
Variable selection can be valuable in the analysis of streaming data with costly measurements, as in intensive care monitoring or battery-powered sensor networks. In the presence of drift, selections must be constantly revised, calling for adaptive variable selection schemes. An important and novel problem arises from the fact that non-selected variables become missing variables, which induces bias upon subsequent decisions. Here, we consider adaptive variable selection in the context of linear regression, using only a fraction of the available regressors per timepoint. We suggest a scheme that fits a multivariate Gaussian over a sliding window using the EM algorithm and selects which variables to observe next using the Lasso algorithm. We experiment with simulated and real data to demonstrate that very high prediction accuracy may be retained using as little as 10% of the data.