Translating feedforward neural nets to SOM-like maps

B.J. van der Zwaag; Lambert Spaanenburg; Cornelis H. Slump

Translating feedforward neural nets to SOM-like maps

B.J. van der Zwaag, Lambert Spaanenburg, Cornelis H. Slump

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

59 Downloads (Pure)

Abstract

A major disadvantage of feedforward neural networks is still the difficulty to gain insight into their internal functionality. This is much less the case for, e.g., nets that are trained unsupervised, such as Kohonen’s self-organizing feature maps (SOMs). These offer a direct view into the stored knowledge, as their internal knowledge is stored in the same format as the input data that was used for training or is used for evaluation. This paper discusses a mathematical transformation of a feed-forward network into a SOMlike structure such that its internal knowledge can be visually interpreted. This is particularly applicable to networks trained in the general classification problem domain.

Original language	Undefined
Title of host publication	14th Annual Workshop on Circuits Systems and Signal Processing (ProRISC)
Place of Publication	Netherlands
Publisher	STW
Pages	447-452
Number of pages	6
ISBN (Print)	90-73461-39-1
Publication status	Published - Nov 2003
Event	14th ProRISC Workshop on Circuits, Systems and Signal Processing 2003 - Veldhoven, Netherlands Duration: 25 Nov 2003 → 27 Nov 2003 Conference number: 14

Publication series

Name
Publisher	STW Technology Foundation

Workshop

Workshop	14th ProRISC Workshop on Circuits, Systems and Signal Processing 2003
Abbreviated title	ProRISC
Country/Territory	Netherlands
City	Veldhoven
Period	25/11/03 → 27/11/03

Keywords

feature maps
selforganizing maps
character recognition
rule extraction
EWI-9666
Neural Networks
IR-46698
METIS-215828

Access to Document

Cite this

@inproceedings{2da7fd09aeb646cc8b7e4783b4a35a84,

title = "Translating feedforward neural nets to SOM-like maps",

abstract = "A major disadvantage of feedforward neural networks is still the difficulty to gain insight into their internal functionality. This is much less the case for, e.g., nets that are trained unsupervised, such as Kohonen{\textquoteright}s self-organizing feature maps (SOMs). These offer a direct view into the stored knowledge, as their internal knowledge is stored in the same format as the input data that was used for training or is used for evaluation. This paper discusses a mathematical transformation of a feed-forward network into a SOMlike structure such that its internal knowledge can be visually interpreted. This is particularly applicable to networks trained in the general classification problem domain.",

keywords = "feature maps, selforganizing maps, character recognition, rule extraction, EWI-9666, Neural Networks, IR-46698, METIS-215828",

author = "{van der Zwaag}, B.J. and Lambert Spaanenburg and Slump, {Cornelis H.}",

note = "SAS 38-03 ; 14th ProRISC Workshop on Circuits, Systems and Signal Processing 2003 ; Conference date: 25-11-2003 Through 27-11-2003",

year = "2003",

month = nov,

language = "Undefined",

isbn = "90-73461-39-1",

publisher = "STW",

pages = "447--452",

booktitle = "14th Annual Workshop on Circuits Systems and Signal Processing (ProRISC)",

}

TY - GEN

T1 - Translating feedforward neural nets to SOM-like maps

AU - van der Zwaag, B.J.

AU - Spaanenburg, Lambert

AU - Slump, Cornelis H.

N1 - Conference code: 14

PY - 2003/11

Y1 - 2003/11

N2 - A major disadvantage of feedforward neural networks is still the difficulty to gain insight into their internal functionality. This is much less the case for, e.g., nets that are trained unsupervised, such as Kohonen’s self-organizing feature maps (SOMs). These offer a direct view into the stored knowledge, as their internal knowledge is stored in the same format as the input data that was used for training or is used for evaluation. This paper discusses a mathematical transformation of a feed-forward network into a SOMlike structure such that its internal knowledge can be visually interpreted. This is particularly applicable to networks trained in the general classification problem domain.

AB - A major disadvantage of feedforward neural networks is still the difficulty to gain insight into their internal functionality. This is much less the case for, e.g., nets that are trained unsupervised, such as Kohonen’s self-organizing feature maps (SOMs). These offer a direct view into the stored knowledge, as their internal knowledge is stored in the same format as the input data that was used for training or is used for evaluation. This paper discusses a mathematical transformation of a feed-forward network into a SOMlike structure such that its internal knowledge can be visually interpreted. This is particularly applicable to networks trained in the general classification problem domain.

KW - feature maps

KW - selforganizing maps

KW - character recognition

KW - rule extraction

KW - EWI-9666

KW - Neural Networks

KW - IR-46698

KW - METIS-215828

M3 - Conference contribution

SN - 90-73461-39-1

SP - 447

EP - 452

BT - 14th Annual Workshop on Circuits Systems and Signal Processing (ProRISC)

PB - STW

CY - Netherlands

T2 - 14th ProRISC Workshop on Circuits, Systems and Signal Processing 2003

Y2 - 25 November 2003 through 27 November 2003

ER -