References & Citations
Statistics > Machine Learning
Title: Variational Joint Filtering
(Submitted on 27 Jul 2017 (v1), revised 13 Nov 2019 (this version, v4), latest version 29 Jun 2020 (v5))
Abstract: New technologies for recording the activity of large neural populations during complex behavior provide exciting opportunities for investigating the neural computations that underlie perception, cognition, and decision-making. Nonlinear state-space models provide an interpretable signal processing framework by combining an intuitive dynamical system with a probabilistic observation model, which can provide insights into neural dynamics, neural computation, and development of neural prosthetics and treatment through feedback control. It yet brings the challenge of learning both latent neural state and the underlying dynamical system because neither is known for neural systems a priori. We developed a flexible online learning framework for latent nonlinear state dynamics and filtered latent states. Using the stochastic gradient variational Bayes approach, our method jointly optimizes the parameters of the nonlinear dynamical system, the observation model, and the black-box recognition model. Unlike previous approaches, our framework can incorporate non-trivial distributions of observation noise and has constant time and space complexity. These features make our approach amenable to real-time applications and the potential to automate analysis and experimental design in ways that testably track and modify behavior using stimuli designed to influence learning.
Submission history
From: Yuan Zhao [view email][v1] Thu, 27 Jul 2017 21:15:30 GMT (3746kb,D)
[v2] Mon, 26 Feb 2018 15:56:20 GMT (3663kb,D)
[v3] Mon, 21 May 2018 15:06:52 GMT (5685kb,D)
[v4] Wed, 13 Nov 2019 20:57:28 GMT (4363kb,D)
[v5] Mon, 29 Jun 2020 14:37:40 GMT (3650kb,D)
Link back to: arXiv, form interface, contact.