Event

Ba-GP Corrects Gaussian Process Input Bias

1 reports · 1 sources

Muzaffar Qureshi and co-authors propose the Bias-Aware Gaussian Process (Ba-GP) framework to correct input bias in Gaussian process regression caused by pose estimation errors of mobile sensors. The method uses the recorded gradient of the GP posterior mean to propagate the estimated input bias and noise into mean and covariance corrections, and provides a first-order correction algorithm based on an additive covariance structure. The authors report that simulation studies and an indoor light-intensity mapping experiment show that Ba-GP improves prediction accuracy and uncertainty calibration relative to a noisy-input GP baseline, at a computational cost an order of magnitude lower than fully retraining the GP. These results come from experiments reported by the authors and await independent verification.