Sparseness-Optimized Feature Importance

Isel Grau; Gonzalo Nápoles

doi:10.1007/978-3-031-63797-1_20

Sparseness-Optimized Feature Importance

Isel Grau
, Gonzalo Nápoles

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

5 Citations (Scopus)

121 Downloads (Pure)

Abstract

In this paper, we propose a model-agnostic post-hoc explanation procedure devoted to computing feature attribution. The proposed method, termed Sparseness-Optimized Feature Importance (SOFI), entails solving an optimization problem related to the sparseness of feature importance explanations. The intuition behind this property is that the model's performance is severely affected after marginalizing the most important features while remaining largely unaffected after marginalizing the least important ones. Existing post-hoc feature attribution methods do not optimize this property directly but rather implement proxies to obtain this behavior. Numerical simulations using both structured (tabular) and unstructured (image) classification datasets show the superiority of our proposal compared with state-of-the-art feature attribution explanation methods. The implementation of the method is available on https://github.com/igraugar/sofi.

Original language	English
Title of host publication	Explainable Artificial Intelligence
Subtitle of host publication	Second World Conference, xAI 2024, Valletta, Malta, July 17–19, 2024, Proceedings, Part II
Editors	Luca Longo, Sebastian Lapuschkin, Christin Seifert
Place of Publication	Cham
Publisher	Springer
Pages	393-415
Number of pages	23
ISBN (Electronic)	978-3-031-63797-1
ISBN (Print)	978-3-031-63796-4
DOIs	https://doi.org/10.1007/978-3-031-63797-1_20
Publication status	Published - 10 Jul 2024
Event	2nd World Conference on eXplainable Artificial Intelligence, xAI 2024 - Valletta, Malta Duration: 17 Jul 2024 → 19 Jul 2024

Publication series

Name	Communications in Computer and Information Science (CCIS)
Volume	2154
ISSN (Print)	1865-0929
ISSN (Electronic)	1865-0937

Conference

Conference	2nd World Conference on eXplainable Artificial Intelligence, xAI 2024
Abbreviated title	xAI 2024
Country/Territory	Malta
City	Valletta
Period	17/07/24 → 19/07/24

Funding

This paper is partially supported by the European Union’s HORIZON Research and Innovation Programme under grant agreement No 101120657, project ENFIELD (European Lighthouse to Manifest Trustworthy and Green AI).

Funders	Funder number
European Union's Horizon 2020 - Research and Innovation Framework Programme	101120657

Keywords

feature importance
model-agnostic explainability
sparse explanations

Access to Document

10.1007/978-3-031-63797-1_20

SOFI_paper_XAI_conference_formatFinal author version , 44.1 MBLicence: TAVERNE
pdfFinal published version, 9.72 MBLicence: TAVERNE

Cite this

Grau, I., & Nápoles, G. (2024). Sparseness-Optimized Feature Importance. In L. Longo, S. Lapuschkin, & C. Seifert (Eds.), Explainable Artificial Intelligence: Second World Conference, xAI 2024, Valletta, Malta, July 17–19, 2024, Proceedings, Part II (pp. 393-415). (Communications in Computer and Information Science (CCIS); Vol. 2154). Springer. https://doi.org/10.1007/978-3-031-63797-1_20

@inproceedings{add4f9ce84f94bb49c280bbc30e3cef8,

title = "Sparseness-Optimized Feature Importance",

abstract = "In this paper, we propose a model-agnostic post-hoc explanation procedure devoted to computing feature attribution. The proposed method, termed Sparseness-Optimized Feature Importance (SOFI), entails solving an optimization problem related to the sparseness of feature importance explanations. The intuition behind this property is that the model's performance is severely affected after marginalizing the most important features while remaining largely unaffected after marginalizing the least important ones. Existing post-hoc feature attribution methods do not optimize this property directly but rather implement proxies to obtain this behavior. Numerical simulations using both structured (tabular) and unstructured (image) classification datasets show the superiority of our proposal compared with state-of-the-art feature attribution explanation methods. The implementation of the method is available on https://github.com/igraugar/sofi.",

keywords = "feature importance, model-agnostic explainability, sparse explanations",

author = "Isel Grau and Gonzalo N{\'a}poles",

year = "2024",

month = jul,

day = "10",

doi = "10.1007/978-3-031-63797-1\_20",

language = "English",

isbn = "978-3-031-63796-4",

series = "Communications in Computer and Information Science (CCIS)",

publisher = "Springer",

pages = "393--415",

editor = "Luca Longo and Sebastian Lapuschkin and Christin Seifert",

booktitle = "Explainable Artificial Intelligence",

address = "Germany",

note = "2nd World Conference on eXplainable Artificial Intelligence, xAI 2024, xAI 2024 ; Conference date: 17-07-2024 Through 19-07-2024",

}

Grau, I & Nápoles, G 2024, Sparseness-Optimized Feature Importance. in L Longo, S Lapuschkin & C Seifert (eds), Explainable Artificial Intelligence: Second World Conference, xAI 2024, Valletta, Malta, July 17–19, 2024, Proceedings, Part II. Communications in Computer and Information Science (CCIS), vol. 2154, Springer, Cham, pp. 393-415, 2nd World Conference on eXplainable Artificial Intelligence, xAI 2024, Valletta, Malta, 17/07/24. https://doi.org/10.1007/978-3-031-63797-1_20

Sparseness-Optimized Feature Importance. / Grau, Isel; Nápoles, Gonzalo.
Explainable Artificial Intelligence: Second World Conference, xAI 2024, Valletta, Malta, July 17–19, 2024, Proceedings, Part II. ed. / Luca Longo; Sebastian Lapuschkin; Christin Seifert. Cham: Springer, 2024. p. 393-415 (Communications in Computer and Information Science (CCIS); Vol. 2154).

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

TY - GEN

T1 - Sparseness-Optimized Feature Importance

AU - Grau, Isel

AU - Nápoles, Gonzalo

PY - 2024/7/10

Y1 - 2024/7/10

N2 - In this paper, we propose a model-agnostic post-hoc explanation procedure devoted to computing feature attribution. The proposed method, termed Sparseness-Optimized Feature Importance (SOFI), entails solving an optimization problem related to the sparseness of feature importance explanations. The intuition behind this property is that the model's performance is severely affected after marginalizing the most important features while remaining largely unaffected after marginalizing the least important ones. Existing post-hoc feature attribution methods do not optimize this property directly but rather implement proxies to obtain this behavior. Numerical simulations using both structured (tabular) and unstructured (image) classification datasets show the superiority of our proposal compared with state-of-the-art feature attribution explanation methods. The implementation of the method is available on https://github.com/igraugar/sofi.

AB - In this paper, we propose a model-agnostic post-hoc explanation procedure devoted to computing feature attribution. The proposed method, termed Sparseness-Optimized Feature Importance (SOFI), entails solving an optimization problem related to the sparseness of feature importance explanations. The intuition behind this property is that the model's performance is severely affected after marginalizing the most important features while remaining largely unaffected after marginalizing the least important ones. Existing post-hoc feature attribution methods do not optimize this property directly but rather implement proxies to obtain this behavior. Numerical simulations using both structured (tabular) and unstructured (image) classification datasets show the superiority of our proposal compared with state-of-the-art feature attribution explanation methods. The implementation of the method is available on https://github.com/igraugar/sofi.

KW - feature importance

KW - model-agnostic explainability

KW - sparse explanations

UR - https://www.scopus.com/pages/publications/85200681919

U2 - 10.1007/978-3-031-63797-1_20

DO - 10.1007/978-3-031-63797-1_20

M3 - Conference contribution

SN - 978-3-031-63796-4

T3 - Communications in Computer and Information Science (CCIS)

SP - 393

EP - 415

BT - Explainable Artificial Intelligence

A2 - Longo, Luca

A2 - Lapuschkin, Sebastian

A2 - Seifert, Christin

PB - Springer

CY - Cham

T2 - 2nd World Conference on eXplainable Artificial Intelligence, xAI 2024

Y2 - 17 July 2024 through 19 July 2024

ER -

Sparseness-Optimized Feature Importance

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

SOFI: Sparseness-optimized feature importance

SOFI: A Sparseness-Optimized Feature Importance Method

SOFI: A Sparseness-Optimized Feature Importance Method

ENFIELD: European Lighthouse to Manifest Trustworthy and Green AI

sofi-image: Sparseness Optimized Feature Importance for Image Classification

sofi-tabular: Sparseness Optimized Feature Importance