東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Biophysically Interpretable Recurren...

Wang, Yuan.

Linked to FindBook

Google Book

Amazon

博客來

Biophysically Interpretable Recurrent Neural Network for Functional Magnetic Resonance Imaging Analysis.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Biophysically Interpretable Recurrent Neural Network for Functional Magnetic Resonance Imaging Analysis./
Author:	Wang, Yuan.
Published:	Ann Arbor : ProQuest Dissertations & Theses, : 2018,
Description:	108 p.
Notes:	Source: Masters Abstracts International, Volume: 79-12.
Contained By:	Masters Abstracts International79-12.
Subject:	Electrical engineering. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10816127
ISBN:	9780355992205

Biophysically Interpretable Recurrent Neural Network for Functional Magnetic Resonance Imaging Analysis.
Wang, Yuan.

Biophysically Interpretable Recurrent Neural Network for Functional Magnetic Resonance Imaging Analysis. - Ann Arbor : ProQuest Dissertations & Theses, 2018 - 108 p.

Source: Masters Abstracts International, Volume: 79-12.

Thesis (Ph.D.)--New York University Tandon School of Engineering, 2018.

This item must not be sold to any third party vendors.

Recently, state-of-the-art Recurrent Neural Networks (RNNs) have been used to expand people's knowledge of neuroscience. However, these generic RNNs lack biophysical meaning, making the interpretation of results in a neuroscience context difficult. In this study, we propose a new biophysically plausible RNN built on Dynamic Causal Modelling (DCM). DCM is a nonlinear generative model explicitly describing the entire process from stimulus to functional magnetic resonance image (fMRI) blood oxygen level dependent (BOLD) signal via neural activity and cerebral hemodynamics, and thus is considered by many to be the most biologically plausible as well as the most technically advanced fMRI modeling method. We propose a generalization of RNN and show that DCM can be cast faithfully as a special form of the generalized RNN. We demonstrate that the maximum a posteriori (MAP) estimation of DCM parameters can be found with back propagation, a novel form of DCM model inversion. DCM-RNN combines biophysical interpretability and neural network compatibility, making it a versatile tool for neuroscience. A combination of DCM-RNN and other deep neural networks enables one to study human brains while subjects perform complex tasks such as reading and watching movies. DCM is traditionally used to estimate the brain effective connectivity, the coupling strength between brain regions, and the brain causal architecture. In experiments of inferring the effective connectivity, back propagation used in DCM-RNN was shown to be more noise robust than the variational Gauss-Newton search traditionally used for DCM parameter estimation. In experiments of model discovering, DCM-RNN, equipped with a l1 prior, was shown to be better at identifying parsimonious causal architecture than the traditional DCM parameter estimation method. We also evaluated the incorporation of a neuron firing model in the DCM-RNN framework. Studies with complex tasks are in our further plan.

ISBN: 9780355992205Subjects--Topical Terms:

649834
Electrical engineering.

Biophysically Interpretable Recurrent Neural Network for Functional Magnetic Resonance Imaging Analysis.
LDR:03040nmm a2200313 4500 001 2211048
005 20191126114025.5
008 201008s2018 ||||||||||||||||| ||eng d
020 $a 9780355992205
035 $a (MiAaPQ)AAI10816127
035 $a (MiAaPQ)poly:10392
035 $a AAI10816127
040 $a MiAaPQ $c MiAaPQ
100 1 $a Wang, Yuan. $3 1273603
245 1 0 $a Biophysically Interpretable Recurrent Neural Network for Functional Magnetic Resonance Imaging Analysis.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2018
300 $a 108 p.
500 $a Source: Masters Abstracts International, Volume: 79-12.
500 $a Publisher info.: Dissertation/Thesis.
500 $a Advisor: Wang, Yao.
502 $a Thesis (Ph.D.)--New York University Tandon School of Engineering, 2018.
506 $a This item must not be sold to any third party vendors.
520 $a Recently, state-of-the-art Recurrent Neural Networks (RNNs) have been used to expand people's knowledge of neuroscience. However, these generic RNNs lack biophysical meaning, making the interpretation of results in a neuroscience context difficult. In this study, we propose a new biophysically plausible RNN built on Dynamic Causal Modelling (DCM). DCM is a nonlinear generative model explicitly describing the entire process from stimulus to functional magnetic resonance image (fMRI) blood oxygen level dependent (BOLD) signal via neural activity and cerebral hemodynamics, and thus is considered by many to be the most biologically plausible as well as the most technically advanced fMRI modeling method. We propose a generalization of RNN and show that DCM can be cast faithfully as a special form of the generalized RNN. We demonstrate that the maximum a posteriori (MAP) estimation of DCM parameters can be found with back propagation, a novel form of DCM model inversion. DCM-RNN combines biophysical interpretability and neural network compatibility, making it a versatile tool for neuroscience. A combination of DCM-RNN and other deep neural networks enables one to study human brains while subjects perform complex tasks such as reading and watching movies. DCM is traditionally used to estimate the brain effective connectivity, the coupling strength between brain regions, and the brain causal architecture. In experiments of inferring the effective connectivity, back propagation used in DCM-RNN was shown to be more noise robust than the variational Gauss-Newton search traditionally used for DCM parameter estimation. In experiments of model discovering, DCM-RNN, equipped with a l1 prior, was shown to be better at identifying parsimonious causal architecture than the traditional DCM parameter estimation method. We also evaluated the incorporation of a neuron firing model in the DCM-RNN framework. Studies with complex tasks are in our further plan.
590 $a School code: 1988.
650 4 $a Electrical engineering. $3 649834
690 $a 0544
710 2 $a New York University Tandon School of Engineering. $b Electrical and Computer Engineering. $3 3350298
773 0 $t Masters Abstracts International $g 79-12.
790 $a 1988
791 $a Ph.D.
792 $a 2018
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10816127