Accepted author manuscript, 522 KB, PDF document
Available under license: Unspecified
Rights statement: Please note: this is the Appendix to the article.
Other version, 416 KB, PDF document
Final published version
Licence: Unspecified
Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSN › Conference contribution/Paper › peer-review
Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSN › Conference contribution/Paper › peer-review
}
TY - GEN
T1 - Recurrent Kalman networks
T2 - factorized inference in high-dimensional deep feature spaces
AU - Becker, P.
AU - Pandya, H.
AU - Gebhardt, G.
AU - Zhao, C.
AU - Taylor, C. James
AU - Neumann, G.
PY - 2019/6/13
Y1 - 2019/6/13
N2 - In order to integrate uncertainty estimates into deep time-series modelling, Kalman Filters (KFs) have been integrated with deep learning models; however, such approaches typically rely on approximate inference techniques such as variational inference which makes learning more complex and often less scalable due to approximation errors. We propose a new deep approach to Kalman filtering which can be learned directly in an end-to-end manner using backpropagation without additional approximations. Our approach uses a high-dimensional factorized latent state representation for which the Kalman updates simplify to scalar operations and thus avoids hard to backpropagate, computationally heavy and potentially unstable matrix inversions. Moreover, we use locally linear dynamic models to efficiently propagate the latent state to the next time step. The resulting network architecture, which we call Recurrent Kalman Network (RKN), can be used for any time-series data, similar to a LSTM (Hochreiter & Schmidhuber, 1997) but uses an explicit representation of uncertainty. As shown by our experiments, the RKN obtains much more accurate uncertainty estimates than an LSTM or Gated Recurrent Units (GRUs) (Cho et al., 2014) while also showing a slightly improved prediction performance and outperforms various recent generative models on an image imputation task.
AB - In order to integrate uncertainty estimates into deep time-series modelling, Kalman Filters (KFs) have been integrated with deep learning models; however, such approaches typically rely on approximate inference techniques such as variational inference which makes learning more complex and often less scalable due to approximation errors. We propose a new deep approach to Kalman filtering which can be learned directly in an end-to-end manner using backpropagation without additional approximations. Our approach uses a high-dimensional factorized latent state representation for which the Kalman updates simplify to scalar operations and thus avoids hard to backpropagate, computationally heavy and potentially unstable matrix inversions. Moreover, we use locally linear dynamic models to efficiently propagate the latent state to the next time step. The resulting network architecture, which we call Recurrent Kalman Network (RKN), can be used for any time-series data, similar to a LSTM (Hochreiter & Schmidhuber, 1997) but uses an explicit representation of uncertainty. As shown by our experiments, the RKN obtains much more accurate uncertainty estimates than an LSTM or Gated Recurrent Units (GRUs) (Cho et al., 2014) while also showing a slightly improved prediction performance and outperforms various recent generative models on an image imputation task.
KW - Kalman Filter
KW - state estimation
KW - robot dynamics
M3 - Conference contribution/Paper
VL - 97
T3 - Proceedings of Machine Learning Research (PMLR)
SP - 544
EP - 552
BT - Proceedings of Machine Learning Research (PMLR)
ER -