Active Monitoring of Neural Networks

A. Lukina; Christian  Schilling; Thomas A.  Henzinger

Active Monitoring of Neural Networks

A. Lukina, Christian Schilling, Thomas A. Henzinger

Algorithmics

Research output: Chapter in Book/Conference proceedings/Edited volume › Conference contribution › Scientific › peer-review

34 Downloads (Pure)

Abstract

Neural-network classifiers are trained to achieve high prediction accuracy. However, their performance still suffers from frequently appearing inputs of unknown classes. As a component of a cyber-physical system, the classifier in this case can no longer be reliable and is typically retrained. We propose an algorithmic framework for monitoring reliability of a neural network. In contrast to static detection, a monitor wrapped in our framework operates in parallel with the classifier, communicates interpretable labeling queries to the human user, and incrementally adapts to their feedback.

Original language	English
Title of host publication	BNAIC/BeneLearn 2021
Subtitle of host publication	33rd Benelux Conference on Artificial Intelligence and 30th Belgian-Dutch Conference on Machine Learning
Editors	Edit Luis A. Leiva, Cédric Pruski, Réka Markovich, Amro Najjar, Christoph Schommer
Pages	685-687
Publication status	Published - 2021
Event	33rd Benelux Conference on Artificial Intelligence and 30th Belgian-Dutch Conference on Machine Learning - Esch-sur-Alzette, Luxembourg Duration: 10 Nov 2021 → 12 Nov 2021

Conference

Conference	33rd Benelux Conference on Artificial Intelligence and 30th Belgian-Dutch Conference on Machine Learning
Abbreviated title	BNAIC/BeneLearn 2021
Country/Territory	Luxembourg
City	Esch-sur-Alzette
Period	10/11/21 → 12/11/21

Keywords

monitoring
neural networks
novelty detection

Access to Document

bnaic2021_preproceedings8Final published version, 56.8 KB

Cite this

@inproceedings{f468eb5188f24f21b928b9516e93e946,

title = "Active Monitoring of Neural Networks",

abstract = "Neural-network classifiers are trained to achieve high prediction accuracy. However, their performance still suffers from frequently appearing inputs of unknown classes. As a component of a cyber-physical system, the classifier in this case can no longer be reliable and is typically retrained. We propose an algorithmic framework for monitoring reliability of a neural network. In contrast to static detection, a monitor wrapped in our framework operates in parallel with the classifier, communicates interpretable labeling queries to the human user, and incrementally adapts to their feedback. ",

keywords = "monitoring, neural networks, novelty detection",

author = "A. Lukina and Christian Schilling and Henzinger, {Thomas A.}",

year = "2021",

language = "English",

pages = "685--687",

editor = "Leiva, {Edit Luis A. } and Pruski, {C{\'e}dric } and Markovich, {R{\'e}ka } and Najjar, {Amro } and Schommer, {Christoph }",

booktitle = "BNAIC/BeneLearn 2021",

note = "33rd Benelux Conference on Artificial Intelligence and<br/>30th Belgian-Dutch Conference on Machine Learning, BNAIC/BeneLearn 2021 ; Conference date: 10-11-2021 Through 12-11-2021",

}

Lukina, A, Schilling, C & Henzinger, TA 2021, Active Monitoring of Neural Networks. in ELA Leiva, C Pruski, R Markovich, A Najjar & C Schommer (eds), BNAIC/BeneLearn 2021: 33rd Benelux Conference on Artificial Intelligence and 30th Belgian-Dutch Conference on Machine Learning. pp. 685-687, 33rd Benelux Conference on Artificial Intelligence and
30th Belgian-Dutch Conference on Machine Learning, Esch-sur-Alzette, Luxembourg, 10/11/21.

Active Monitoring of Neural Networks. / Lukina, A.; Schilling, Christian ; Henzinger, Thomas A. .
BNAIC/BeneLearn 2021: 33rd Benelux Conference on Artificial Intelligence and 30th Belgian-Dutch Conference on Machine Learning. ed. / Edit Luis A. Leiva; Cédric Pruski; Réka Markovich; Amro Najjar; Christoph Schommer. 2021. p. 685-687.

Research output: Chapter in Book/Conference proceedings/Edited volume › Conference contribution › Scientific › peer-review

TY - GEN

T1 - Active Monitoring of Neural Networks

AU - Lukina, A.

AU - Schilling, Christian

AU - Henzinger, Thomas A.

PY - 2021

Y1 - 2021

N2 - Neural-network classifiers are trained to achieve high prediction accuracy. However, their performance still suffers from frequently appearing inputs of unknown classes. As a component of a cyber-physical system, the classifier in this case can no longer be reliable and is typically retrained. We propose an algorithmic framework for monitoring reliability of a neural network. In contrast to static detection, a monitor wrapped in our framework operates in parallel with the classifier, communicates interpretable labeling queries to the human user, and incrementally adapts to their feedback.

AB - Neural-network classifiers are trained to achieve high prediction accuracy. However, their performance still suffers from frequently appearing inputs of unknown classes. As a component of a cyber-physical system, the classifier in this case can no longer be reliable and is typically retrained. We propose an algorithmic framework for monitoring reliability of a neural network. In contrast to static detection, a monitor wrapped in our framework operates in parallel with the classifier, communicates interpretable labeling queries to the human user, and incrementally adapts to their feedback.

KW - monitoring

KW - neural networks

KW - novelty detection

M3 - Conference contribution

SP - 685

EP - 687

BT - BNAIC/BeneLearn 2021

A2 - Leiva, Edit Luis A.

A2 - Pruski, Cédric

A2 - Markovich, Réka

A2 - Najjar, Amro

A2 - Schommer, Christoph

T2 - 33rd Benelux Conference on Artificial Intelligence and<br/>30th Belgian-Dutch Conference on Machine Learning

Y2 - 10 November 2021 through 12 November 2021

ER -

Active Monitoring of Neural Networks

Abstract

Conference

Keywords

Access to Document

Fingerprint

Cite this