Structured Variational Inference in Unstable Gaussian Process State Space Models
arXiv:1907.07035
Abstract
We propose a new variational inference algorithm for learning in Gaussian Process State-Space Models (GPSSMs). Our algorithm enables learning of unstable and partially observable systems, where previous algorithms fail. Our main algorithmic contribution is a novel approximate posterior that can be calculated efficiently using a single forward and backward pass along the training trajectories. The forward-backward pass is inspired on Kalman smoothing for linear dynamical systems but generalizes to GPSSMs. Our second contribution is a modification of the conditioning step that effectively lowers the Kalman gain. This modification is crucial to attaining good test performance where no measurements are available. Finally, we show experimentally that our learning algorithm performs well in stable and unstable real systems with hidden states.
References in corpus (7)
- Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation
- On Calibration of Modern Neural Networks
- Data-Efficient Reinforcement Learning with Probabilistic Model Predictive Control
- Variational Gaussian Process State-Space Models
- Black box variational inference for state space models
- Recurrent Gaussian Processes
- Non-Factorised Variational Inference in Dynamical Systems