MLLGDec 14, 2018

Non-Factorised Variational Inference in Dynamical Systems

arXiv:1812.06067v16 citations
Originality Incremental advance
AI Analysis

This work addresses a specific issue in probabilistic modeling of dynamical systems, offering an incremental improvement over existing factorised approaches.

The paper tackles the problem of overconfident posterior estimates and overestimated process noise in variational inference for dynamical systems with Gaussian process transition functions, by introducing a non-factorised posterior method that maintains computational efficiency.

We focus on variational inference in dynamical systems where the discrete time transition function (or evolution rule) is modelled by a Gaussian process. The dominant approach so far has been to use a factorised posterior distribution, decoupling the transition function from the system states. This is not exact in general and can lead to an overconfident posterior over the transition function as well as an overestimation of the intrinsic stochasticity of the system (process noise). We propose a new method that addresses these issues and incurs no additional computational costs.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes