Efficient learning of nonlinear prediction models with time-series privileged information
This work addresses the challenge of limited sample sizes in domains like dynamical systems, offering a method to enhance prediction efficiency, though it builds incrementally on prior linear-Gaussian results.
The paper tackles the problem of improving sample efficiency in nonlinear prediction tasks by extending learning using privileged information (LuPI) to time-series data, showing that LuPI learners are never worse and often better than classical learners, with empirical results confirming these theoretical guarantees.
In domains where sample sizes are limited, efficient learning algorithms are critical. Learning using privileged information (LuPI) offers increased sample efficiency by allowing prediction models access to auxiliary information at training time which is unavailable when the models are used. In recent work, it was shown that for prediction in linear-Gaussian dynamical systems, a LuPI learner with access to intermediate time series data is never worse and often better in expectation than any unbiased classical learner. We provide new insights into this analysis and generalize it to nonlinear prediction tasks in latent dynamical systems, extending theoretical guarantees to the case where the map connecting latent variables and observations is known up to a linear transform. In addition, we propose algorithms based on random features and representation learning for the case when this map is unknown. A suite of empirical results confirm theoretical findings and show the potential of using privileged time-series information in nonlinear prediction.