Paper Abstract and Keywords |
Presentation |
2017-01-26 16:00
Fast Receptive field Inference with Sparse Fourirer Representation by using LASSO Takeshi Tanida, Hirotaka Sakamoto, Yasuhiko Igarashi, Takeshi Ideriha, Satoru Tokuda (Univ. of Tokyo), Kota Sasaki, Izumi Ohzawa (Osaka Univ.), Masato Okada (Univ. of Tokyo/RIKEN) NC2016-52 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
We propose fast receptive eld(RF) inference. The RF describes how a neuron sums up its inputs across
space and time. The traditional RF estimators such as the spike-triggered average, converge slowly and often require
large amounts of spike data. Previous research introduce a family of prior distribution to low cost estimation, by
utilizing an approach known as empirical Bayes. In this study, we estimate the accurate RF by using regression
analysis and variable selection based on the least absolute shrinkage and selection operator (Lasso) with respect to
the Fourier coefficients of the STA data. On the assumption that the RF has sparsity in the Fourier representation,
the Lasso gives the denoised RF estimator. We compare our proposed method with the previous Bayesian methods,
in the experiments of RF estimation by using articial and real data sets. We show that our method is robuster
than the previous method and can estimate fast and accurately, in the case that the observed spike data are few. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Receptive Field Inference, / Spike Triggered Average, / Lasso / / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 116, no. 424, NC2016-52, pp. 25-30, Jan. 2017. |
Paper # |
NC2016-52 |
Date of Issue |
2017-01-19 (NC) |
ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
NC2016-52 |
|