LEMON: Alternative Sampling for More Faithful Explanation Through Local Surrogate Models

Dennis Collaris; Pratik Gajane; Joost Jorritsma; Jack J. van Wijk; Mykola Pechenizkiy

doi:10.1007/978-3-031-30047-9_7

LEMON: Alternative Sampling for More Faithful Explanation Through Local Surrogate Models

Dennis Collaris (Corresponding author), Pratik Gajane, Joost Jorritsma, Jack J. van Wijk, Mykola Pechenizkiy

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

5 Citations (Scopus)

6 Downloads (Pure)

Abstract

Local surrogate learning is a popular and successful method for machine learning explanation. It uses synthetic transfer data to approximate a complex reference model. The sampling technique used for this transfer data has a significant impact on the provided explanation, but remains relatively unexplored in literature. In this work, we explore alternative sampling techniques in pursuit of more faithful and robust explanations, and present LEMON: a sampling technique that samples directly from the desired distribution instead of reweighting samples as done in other explanation techniques (e.g., LIME). Next, we evaluate our technique in a synthetic and UCI dataset-based experiment, and show that our sampling technique yields more faithful explanations compared to current state-of-the-art explainers.

Original language	English
Title of host publication	Advances in Intelligent Data Analysis XXI
Subtitle of host publication	21st International Symposium on Intelligent Data Analysis, IDA 2023, Louvain-la-Neuve, Belgium, April 12–14, 2023, Proceedings
Editors	Bruno Crémilleux, Sibylle Hess, Siegfried Nijssen
Publisher	Springer
Pages	77-90
Number of pages	14
ISBN (Electronic)	978-3-031-30047-9
ISBN (Print)	978-3-031-30046-2
DOIs	https://doi.org/10.1007/978-3-031-30047-9_7
Publication status	Published - 1 Apr 2023
Event	21st International Symposium on Intelligent Data Analysis - Louvain-la-Neuve, Belgium Duration: 12 Apr 2023 → 14 Apr 2023 https://ida2023.org

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13876 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	21st International Symposium on Intelligent Data Analysis
Abbreviated title	IDA 2023
Country/Territory	Belgium
City	Louvain-la-Neuve
Period	12/04/23 → 14/04/23
Internet address	https://ida2023.org

Funding

Acknowledgments. This work is part of the TEPAIV research project with project number 612.001.752, the NWO research project with project number 613.009.122, and the research programme Commit2Data, specifically the RATE Analytics project with project number 628.003.001, which are all financed by the Dutch Research Council (NWO).

Funders	Funder number
Nederlandse Organisatie voor Wetenschappelijk Onderzoek	613.009.122, 628.003.001

Keywords

Machine learning
Interpretability
XAI
LIME
Sampling
Explainable AI

Access to Document

10.1007/978-3-031-30047-9_7

pdfFinal published version, 988 KBLicence: CC BY

Cite this

Collaris, D., Gajane, P., Jorritsma, J., van Wijk, J. J., & Pechenizkiy, M. (2023). LEMON: Alternative Sampling for More Faithful Explanation Through Local Surrogate Models. In B. Crémilleux, S. Hess, & S. Nijssen (Eds.), Advances in Intelligent Data Analysis XXI: 21st International Symposium on Intelligent Data Analysis, IDA 2023, Louvain-la-Neuve, Belgium, April 12–14, 2023, Proceedings (pp. 77-90). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13876 LNCS). Springer. https://doi.org/10.1007/978-3-031-30047-9_7

Collaris, Dennis ; Gajane, Pratik ; Jorritsma, Joost et al. / LEMON : Alternative Sampling for More Faithful Explanation Through Local Surrogate Models. Advances in Intelligent Data Analysis XXI: 21st International Symposium on Intelligent Data Analysis, IDA 2023, Louvain-la-Neuve, Belgium, April 12–14, 2023, Proceedings. editor / Bruno Crémilleux ; Sibylle Hess ; Siegfried Nijssen. Springer, 2023. pp. 77-90 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{0c37324f85f8466fb532d7d9557ccae7,

title = "LEMON: Alternative Sampling for More Faithful Explanation Through Local Surrogate Models",

abstract = "Local surrogate learning is a popular and successful method for machine learning explanation. It uses synthetic transfer data to approximate a complex reference model. The sampling technique used for this transfer data has a significant impact on the provided explanation, but remains relatively unexplored in literature. In this work, we explore alternative sampling techniques in pursuit of more faithful and robust explanations, and present LEMON: a sampling technique that samples directly from the desired distribution instead of reweighting samples as done in other explanation techniques (e.g., LIME). Next, we evaluate our technique in a synthetic and UCI dataset-based experiment, and show that our sampling technique yields more faithful explanations compared to current state-of-the-art explainers.",

keywords = "Machine learning, Interpretability, XAI, LIME, Sampling, Explainable AI",

author = "Dennis Collaris and Pratik Gajane and Joost Jorritsma and \{van Wijk\}, \{Jack J.\} and Mykola Pechenizkiy",

year = "2023",

month = apr,

day = "1",

doi = "10.1007/978-3-031-30047-9\_7",

language = "English",

isbn = "978-3-031-30046-2",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer",

pages = "77--90",

editor = "Bruno Cr{\'e}milleux and Sibylle Hess and Siegfried Nijssen",

booktitle = "Advances in Intelligent Data Analysis XXI",

address = "Germany",

note = "21st International Symposium on Intelligent Data Analysis, IDA 2023 ; Conference date: 12-04-2023 Through 14-04-2023",

url = "https://ida2023.org",

}

Collaris, D, Gajane, P, Jorritsma, J, van Wijk, JJ & Pechenizkiy, M 2023, LEMON: Alternative Sampling for More Faithful Explanation Through Local Surrogate Models. in B Crémilleux, S Hess & S Nijssen (eds), Advances in Intelligent Data Analysis XXI: 21st International Symposium on Intelligent Data Analysis, IDA 2023, Louvain-la-Neuve, Belgium, April 12–14, 2023, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13876 LNCS, Springer, pp. 77-90, 21st International Symposium on Intelligent Data Analysis, Louvain-la-Neuve, Belgium, 12/04/23. https://doi.org/10.1007/978-3-031-30047-9_7

LEMON: Alternative Sampling for More Faithful Explanation Through Local Surrogate Models. / Collaris, Dennis (Corresponding author); Gajane, Pratik; Jorritsma, Joost et al.
Advances in Intelligent Data Analysis XXI: 21st International Symposium on Intelligent Data Analysis, IDA 2023, Louvain-la-Neuve, Belgium, April 12–14, 2023, Proceedings. ed. / Bruno Crémilleux; Sibylle Hess; Siegfried Nijssen. Springer, 2023. p. 77-90 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13876 LNCS).

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

TY - GEN

T1 - LEMON

T2 - 21st International Symposium on Intelligent Data Analysis

AU - Collaris, Dennis

AU - Gajane, Pratik

AU - Jorritsma, Joost

AU - van Wijk, Jack J.

AU - Pechenizkiy, Mykola

PY - 2023/4/1

Y1 - 2023/4/1

N2 - Local surrogate learning is a popular and successful method for machine learning explanation. It uses synthetic transfer data to approximate a complex reference model. The sampling technique used for this transfer data has a significant impact on the provided explanation, but remains relatively unexplored in literature. In this work, we explore alternative sampling techniques in pursuit of more faithful and robust explanations, and present LEMON: a sampling technique that samples directly from the desired distribution instead of reweighting samples as done in other explanation techniques (e.g., LIME). Next, we evaluate our technique in a synthetic and UCI dataset-based experiment, and show that our sampling technique yields more faithful explanations compared to current state-of-the-art explainers.

AB - Local surrogate learning is a popular and successful method for machine learning explanation. It uses synthetic transfer data to approximate a complex reference model. The sampling technique used for this transfer data has a significant impact on the provided explanation, but remains relatively unexplored in literature. In this work, we explore alternative sampling techniques in pursuit of more faithful and robust explanations, and present LEMON: a sampling technique that samples directly from the desired distribution instead of reweighting samples as done in other explanation techniques (e.g., LIME). Next, we evaluate our technique in a synthetic and UCI dataset-based experiment, and show that our sampling technique yields more faithful explanations compared to current state-of-the-art explainers.

KW - Machine learning

KW - Interpretability

KW - XAI

KW - LIME

KW - Sampling

KW - Explainable AI

UR - https://explaining.ml/lemon

UR - http://www.scopus.com/inward/record.url?scp=85152543452&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-30047-9_7

DO - 10.1007/978-3-031-30047-9_7

M3 - Conference contribution

SN - 978-3-031-30046-2

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 77

EP - 90

BT - Advances in Intelligent Data Analysis XXI

A2 - Crémilleux, Bruno

A2 - Hess, Sibylle

A2 - Nijssen, Siegfried

PB - Springer

Y2 - 12 April 2023 through 14 April 2023

ER -

Collaris D, Gajane P, Jorritsma J, van Wijk JJ , Pechenizkiy M. LEMON: Alternative Sampling for More Faithful Explanation Through Local Surrogate Models. In Crémilleux B, Hess S, Nijssen S, editors, Advances in Intelligent Data Analysis XXI: 21st International Symposium on Intelligent Data Analysis, IDA 2023, Louvain-la-Neuve, Belgium, April 12–14, 2023, Proceedings. Springer. 2023. p. 77-90. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-30047-9_7

LEMON: Alternative Sampling for More Faithful Explanation Through Local Surrogate Models

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

IDA 2023 Runner-up Frontier Prize

Project LEMON runner-up at IDA Frontier Prize

Cite this

LEMON: Alternative Sampling for More Faithful Explanation Through Local Surrogate Models

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Prizes

IDA 2023 Runner-up Frontier Prize

Press/Media

Project LEMON runner-up at IDA Frontier Prize

Cite this