GP Kernels for Cross-Spectrum Analysis of Dynamic Networks

Abstract

Multi-output Gaussian processes provide a convenient framework for analysis of a network of dynamic systems. For Navy applications,"" such a framework is useful for analysis of multiple streams of data from a network of agents, in which the data may be multi-modal,"" noisy, dynamic, contradictory and sampled non-uniformly in time. Such a model is also applicable to biomedical applications. For ex""ample, when considering multi-region lectrophysiological time-series data, where experimentalists are interested in both power and p""hase coherence between channels. Recently, thespectral mixture (SM) kernel has been proposed to model the spectral density of a sin"gle channel in a Gaussian process (GP) framework. In the proposed research program we will develop a novel covariance kernel for mul"tiple channels, called the cross-spectral mixture (CSM) kernel. This new, flexible kernel represents both the power and phase relati"onship between multiple observation channels. We will demonstrate the expressive capabilities of the CSMkernel through implementati"on of a Bayesian hidden Markov model, where the emission distribution is a multi-output Gaussian process with a CSM covariance kerne"l. Encouraging preliminary and motivating results are presented for measured multi-region electrophysiological data. Within the prop"osed research we will extend this model in two important ways, for applicability to Navy problems: (i) it will be generalized beyond"" the assumption of real-valued data, considering networked data in the form of text, imagery, video and unstructured data; and (ii)"" the GP framework will be generalized via a parametric convolutional filter bank setup, allowing scaling to networks of massive scal"e.

Document Details

Document Type
DoD Grant Award
Publication Date
Sep 01, 2017
Source ID
N000141712841

Entities

People

  • Lawrence Carin

Organizations

  • Duke University
  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Neural Network Machine Learning.
  • Theoretical Analysis.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • Biotechnology
  • Space