Error-adaptive modeling of streaming time-series data using radial basis functions

摘要: Abstract We propose an error-adaptive method for constructing time-series models novelty detection using radial basis functions. The optimization problem is implemented in incremental, or online manner, and permits the balancing of quality data fit sparsity model objective function a linear program. resulting algorithm applied to several examples streaming data. Novel points are identified infeasibility criterion while feasible not used training may be discarded. In practice, only small fraction observed actually required update model. promoting term serves determine location number RBFs Balancing with accuracy allows user adjust complexity. It appears that ability adapt error bound help prevent over-fitting. illustrate approach show fitting procedure robust additive Gaussian noise non-stationary variance. compare proposed other methods literature, including both batch algorithms.