Skip to content

Detail of publication


M. Bulín and L. Šmídl : Insight of Neural Network by Removing Synapses . SVK FAV 2017 – magisterské a doktorské studijní programy, p. 39-40, Zapadočeská univerzita v Plzni, Vladimír Lukeš, Univerzitní 8, 306 14 Plzeň, 2017.

Additional information



Neural networks can be trained to work well for particular tasks, but hardly ever we know why they work so well. Due to the complicated architectures and an enormous number of parameters we usually have a well-working black box and it is hard if not impossible to make targeted changes in a trained model. My work is focused on understanding the behaviour of feedforward neural networks when classifying particular data. The method rests in removing unimportant synapses from a trained network, while the classification accuracy is kept. Based on my experience, over 90% of the synapses are usually redundant in fully-connected networks.

Detail of publication

Title: Insight of Neural Network by Removing Synapses
Author: M. Bulín ; L. Šmídl
Language: English
Date of publication: 25 May 2017
Year: 2017
Type of publication: Papers in proceedings of reviewed conferences
Title of journal or book: SVK FAV 2017 – magisterské a doktorské studijní programy
Page: 39 - 40
ISBN: 978-80-261-0706-4
Editor: Vladimír Lukeš
Publisher: Zapadočeská univerzita v Plzni
Address: Univerzitní 8, 306 14 Plzeň
/ 2019-02-14 10:39:17 /


 author = {M. Bul\'{i}n and L. \v{S}m\'{i}dl},
 title = {Insight of Neural Network by Removing Synapses},
 year = {2017},
 publisher = {Zapado\v{c}esk\'{a} univerzita v Plzni},
 journal = {SVK FAV 2017 - magistersk\'{e} a doktorsk\'{e} studijn\'{i} programy},
 address = {Univerzitn\'{i} 8, 306 14 Plze\v{n}},
 pages = {39-40},
 editor = {Vladim\'{i}r Luke\v{s}},
 ISBN = {978-80-261-0706-4},
 url = {},