Interpretation of Neural Network Players for a Generalized Divide the Dollar Game Using SHAP Values

Garrison W. Greenwood; Hussin Abbass; Aya Hussein

doi:10.1109/SSCI52147.2023.10371984

Interpretation of Neural Network Players for a Generalized Divide the Dollar Game Using SHAP Values

Garrison W. Greenwood, Hussin Abbass, Aya Hussein

Research output: A Conference proceeding or a Chapter in Book › Conference contribution › peer-review

1 Citation (Scopus)

Abstract

Machine learning models can make accurate predictions but trust in the models depends on being able to under-stand why those predictions were made. Unfortunately, machine learning models are black boxes making interpretation difficult. Previously we used an evolutionary algorithm to evolve triplets of neural network players for instances of the Generalized Divide-the-Dollar, which is an economic bargaining game. The players produced fair bids with high bid totals, which is a desirable outcome, but no attempt was made to understand why the players performed so well. In this paper, we interpret the behavior of those neural networks using SHapley Additive exPlanations (or SHAP). Surprisingly, the neural network players exhibited both altruistic and exploitative behavior. Both a global and a local interpretation analysis is presented. The experiments conducted in this work demonstrate a simple method for understanding players' strategies in multi-player gamcs.

Original language	English
Title of host publication	2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023
Editors	Wen Yu, Agung Julius, Minwoo Jake Lee, Dianhui Wang, Zhi-Hui Zhan, Xiaoou Li
Publisher	IEEE, Institute of Electrical and Electronics Engineers
Pages	1808-1813
Number of pages	6
ISBN (Electronic)	9781665430654
DOIs	https://doi.org/10.1109/SSCI52147.2023.10371984
Publication status	Published - Dec 2023
Externally published	Yes
Event	2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023 - Mexico City, Mexico Duration: 5 Dec 2023 → 8 Dec 2023

Publication series

Name	2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023

Conference

Conference	2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023
Country/Territory	Mexico
City	Mexico City
Period	5/12/23 → 8/12/23

Access to Document

10.1109/SSCI52147.2023.10371984

Cite this

Greenwood, G. W., Abbass, H., & Hussein, A. (2023). Interpretation of Neural Network Players for a Generalized Divide the Dollar Game Using SHAP Values. In W. Yu, A. Julius, M. Jake Lee, D. Wang, Z-H. Zhan, & X. Li (Eds.), 2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023 (pp. 1808-1813). (2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023). IEEE, Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/SSCI52147.2023.10371984

Greenwood, Garrison W. ; Abbass, Hussin ; Hussein, Aya. / Interpretation of Neural Network Players for a Generalized Divide the Dollar Game Using SHAP Values. 2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023. editor / Wen Yu ; Agung Julius ; Minwoo Jake Lee ; Dianhui Wang ; Zhi-Hui Zhan ; Xiaoou Li. IEEE, Institute of Electrical and Electronics Engineers, 2023. pp. 1808-1813 (2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023).

@inproceedings{4aa04229a3ef43c89816c6576106f8ee,

title = "Interpretation of Neural Network Players for a Generalized Divide the Dollar Game Using SHAP Values",

abstract = "Machine learning models can make accurate predictions but trust in the models depends on being able to under-stand why those predictions were made. Unfortunately, machine learning models are black boxes making interpretation difficult. Previously we used an evolutionary algorithm to evolve triplets of neural network players for instances of the Generalized Divide-the-Dollar, which is an economic bargaining game. The players produced fair bids with high bid totals, which is a desirable outcome, but no attempt was made to understand why the players performed so well. In this paper, we interpret the behavior of those neural networks using SHapley Additive exPlanations (or SHAP). Surprisingly, the neural network players exhibited both altruistic and exploitative behavior. Both a global and a local interpretation analysis is presented. The experiments conducted in this work demonstrate a simple method for understanding players' strategies in multi-player gamcs.",

keywords = "explainable AI, machine learning, model interpretation",

author = "Greenwood, {Garrison W.} and Hussin Abbass and Aya Hussein",

note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; 2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023 ; Conference date: 05-12-2023 Through 08-12-2023",

year = "2023",

month = dec,

doi = "10.1109/SSCI52147.2023.10371984",

language = "English",

series = "2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023",

publisher = "IEEE, Institute of Electrical and Electronics Engineers",

pages = "1808--1813",

editor = "Wen Yu and Agung Julius and {Jake Lee}, Minwoo and Dianhui Wang and Zhi-Hui Zhan and Xiaoou Li",

booktitle = "2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023",

address = "United States",

}

Greenwood, GW, Abbass, H & Hussein, A 2023, Interpretation of Neural Network Players for a Generalized Divide the Dollar Game Using SHAP Values. in W Yu, A Julius, M Jake Lee, D Wang, Z-H Zhan & X Li (eds), 2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023. 2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023, IEEE, Institute of Electrical and Electronics Engineers, pp. 1808-1813, 2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023, Mexico City, Mexico, 5/12/23. https://doi.org/10.1109/SSCI52147.2023.10371984

Interpretation of Neural Network Players for a Generalized Divide the Dollar Game Using SHAP Values. / Greenwood, Garrison W.; Abbass, Hussin; Hussein, Aya.
2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023. ed. / Wen Yu; Agung Julius; Minwoo Jake Lee; Dianhui Wang; Zhi-Hui Zhan; Xiaoou Li. IEEE, Institute of Electrical and Electronics Engineers, 2023. p. 1808-1813 (2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023).

Research output: A Conference proceeding or a Chapter in Book › Conference contribution › peer-review

TY - GEN

T1 - Interpretation of Neural Network Players for a Generalized Divide the Dollar Game Using SHAP Values

AU - Greenwood, Garrison W.

AU - Abbass, Hussin

AU - Hussein, Aya

PY - 2023/12

Y1 - 2023/12

N2 - Machine learning models can make accurate predictions but trust in the models depends on being able to under-stand why those predictions were made. Unfortunately, machine learning models are black boxes making interpretation difficult. Previously we used an evolutionary algorithm to evolve triplets of neural network players for instances of the Generalized Divide-the-Dollar, which is an economic bargaining game. The players produced fair bids with high bid totals, which is a desirable outcome, but no attempt was made to understand why the players performed so well. In this paper, we interpret the behavior of those neural networks using SHapley Additive exPlanations (or SHAP). Surprisingly, the neural network players exhibited both altruistic and exploitative behavior. Both a global and a local interpretation analysis is presented. The experiments conducted in this work demonstrate a simple method for understanding players' strategies in multi-player gamcs.

AB - Machine learning models can make accurate predictions but trust in the models depends on being able to under-stand why those predictions were made. Unfortunately, machine learning models are black boxes making interpretation difficult. Previously we used an evolutionary algorithm to evolve triplets of neural network players for instances of the Generalized Divide-the-Dollar, which is an economic bargaining game. The players produced fair bids with high bid totals, which is a desirable outcome, but no attempt was made to understand why the players performed so well. In this paper, we interpret the behavior of those neural networks using SHapley Additive exPlanations (or SHAP). Surprisingly, the neural network players exhibited both altruistic and exploitative behavior. Both a global and a local interpretation analysis is presented. The experiments conducted in this work demonstrate a simple method for understanding players' strategies in multi-player gamcs.

KW - explainable AI

KW - machine learning

KW - model interpretation

UR - http://www.scopus.com/inward/record.url?scp=85182920495&partnerID=8YFLogxK

UR - https://conf.papercept.net/conferences/conferences/SSCI23/program/SSCI23_ProgramAtAGlanceWeb.html

UR - https://ieeexplore.ieee.org/xpl/conhome/10371778/proceeding

U2 - 10.1109/SSCI52147.2023.10371984

DO - 10.1109/SSCI52147.2023.10371984

M3 - Conference contribution

AN - SCOPUS:85182920495

T3 - 2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023

SP - 1808

EP - 1813

BT - 2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023

A2 - Yu, Wen

A2 - Julius, Agung

A2 - Jake Lee, Minwoo

A2 - Wang, Dianhui

A2 - Zhan, Zhi-Hui

A2 - Li, Xiaoou

PB - IEEE, Institute of Electrical and Electronics Engineers

T2 - 2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023

Y2 - 5 December 2023 through 8 December 2023

ER -

Greenwood GW, Abbass H, Hussein A. Interpretation of Neural Network Players for a Generalized Divide the Dollar Game Using SHAP Values. In Yu W, Julius A, Jake Lee M, Wang D, Zhan Z-H, Li X, editors, 2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023. IEEE, Institute of Electrical and Electronics Engineers. 2023. p. 1808-1813. (2023 IEEE Symposium Series on Computational Intelligence, SSCI 2023). doi: 10.1109/SSCI52147.2023.10371984

Interpretation of Neural Network Players for a Generalized Divide the Dollar Game Using SHAP Values

Abstract

Publication series

Conference

Access to Document

Other files and links

Fingerprint

Cite this