TU Darmstadt / ULB / TUbiblio

Marginalized Continuous Time Bayesian Networks for Network Reconstruction from Incomplete Observations

Studer, L. and Paulevé, L. and Zechner, C. and Reumann, M. and Rodriguez Martinez, M. and Koeppl, H. (2016):
Marginalized Continuous Time Bayesian Networks for Network Reconstruction from Incomplete Observations.
Phoenix, USA, In: AAAI, Association for the Advancement of Artificial Intelligence, Phoenix, USA, 12.-17.02.2016, [Online-Edition: http://www.aaai.org/ocs/index.php/AAAI/AAAI16/paper/view/123...],
[Conference or Workshop Item]

Abstract

Continuous Time Bayesian Networks (CTBNs) provide a powerful means to model complex network dynamics. However, their inference is computationally demanding—especially if one considers incomplete and noisy time-series data. The latter gives rise to a joint stateand parameter estimation problem, which can only be solved numerically. However, finding the exact parameterization of the CTBN has only secondary importance in most practical scenarios. We therefore propose a method that circumvents the inference of parameters by analytically marginalizing the Markov chain underlying the CTBN model. Since the resulting stochastic process does no longer depend on any model parameters, its inference reduces to a plain nonlinear filtering problem. We solve the latter using a highly efficient implementation of a sequential Monte Carlo scheme. Our framework enables CTBN learning to be applied to incomplete noisy time-series data such as frequently found in molecular biology and other disciplines.

Item Type: Conference or Workshop Item
Erschienen: 2016
Creators: Studer, L. and Paulevé, L. and Zechner, C. and Reumann, M. and Rodriguez Martinez, M. and Koeppl, H.
Title: Marginalized Continuous Time Bayesian Networks for Network Reconstruction from Incomplete Observations
Language: English
Abstract:

Continuous Time Bayesian Networks (CTBNs) provide a powerful means to model complex network dynamics. However, their inference is computationally demanding—especially if one considers incomplete and noisy time-series data. The latter gives rise to a joint stateand parameter estimation problem, which can only be solved numerically. However, finding the exact parameterization of the CTBN has only secondary importance in most practical scenarios. We therefore propose a method that circumvents the inference of parameters by analytically marginalizing the Markov chain underlying the CTBN model. Since the resulting stochastic process does no longer depend on any model parameters, its inference reduces to a plain nonlinear filtering problem. We solve the latter using a highly efficient implementation of a sequential Monte Carlo scheme. Our framework enables CTBN learning to be applied to incomplete noisy time-series data such as frequently found in molecular biology and other disciplines.

Place of Publication: Phoenix, USA
Uncontrolled Keywords: sequential Monte Carlo; graph reconstruction; continuous time Bayesian network
Divisions: 18 Department of Electrical Engineering and Information Technology
18 Department of Electrical Engineering and Information Technology > Institute for Telecommunications > Bioinspired Communication Systems
18 Department of Electrical Engineering and Information Technology > Institute for Telecommunications
Event Title: AAAI, Association for the Advancement of Artificial Intelligence
Event Location: Phoenix, USA
Event Dates: 12.-17.02.2016
Date Deposited: 08 Dec 2015 10:59
Official URL: http://www.aaai.org/ocs/index.php/AAAI/AAAI16/paper/view/123...
Additional Information:

Copyright 2015 Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.

Export:

Optionen (nur für Redakteure)

View Item View Item