Efficient learning with privileged information in nonlinear time series

Typ
Examensarbete för masterexamen
Program
Publicerad
2022
Författare
Jung, Bastian
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
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 information at training time that 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 timeseries information in nonlinear prediction.
Beskrivning
Ämne/nyckelord
Machine Learning, Privileged Information, Time Series, Sample Efficiency, Latent Dynamical Systems
Citation
Arkitekt (konstruktör)
Geografisk plats
Byggnad (typ)
Byggår
Modelltyp
Skala
Teknik / material