Interpreting Tactical Decision-making in Transformer-based Agents

Koen Boeckx; Xavier Neyt

Interpreting Tactical Decision-making in Transformer-based Agents

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

7 Downloads (Pure)

Abstract

The development of autonomous agents capable of sophisticated strategic decision-making in complex
environments is a central goal of artificial intelligence. This paper proposes a framework for discovering and interpreting
strategies in a simulated grid-world battlefield environment. We leverage the AlphaZero algorithm, a powerful reinforcement
learning method that combines Monte Carlo Tree Search (MCTS) with deep neural networks, to train agents. Crucially, the
neural network component incorporates a Transformer architecture. The primary contribution of this work lies in the proposed
methodology to utilize the self-attention mechanisms within the Transformer to gain insights into the agent’s decision-making
process, specifically by visualizing which parts of the battlefield the network pays attention to when selecting an action. This

Original language	English
Title of host publication	Proceedings of the 1st International Conference on Big Data & Applications
Subtitle of host publication	BDAA 2025
Editors	Sergey Y. Yurish
Place of Publication	2025
Publisher	International Frequency Sensor Association (IFSA) Publishing
Pages	51-55
Number of pages	5
ISBN (Electronic)	978-84-09-78845-3
Publication status	Published - 25 Nov 2025
Event	1st International Conference on Big Data Analytics and Applications - Austria Trend Hotel Congress Innsbruck, Innsbruck, Austria Duration: 25 Nov 2025 → 27 Nov 2025 Conference number: 1 https://bdaa-conference.com/

Conference

Conference	1st International Conference on Big Data Analytics and Applications
Abbreviated title	BDAA 2025
Country/Territory	Austria
City	Innsbruck
Period	25/11/25 → 27/11/25
Internet address	https://bdaa-conference.com/

Access to Document

18_Paper_for_BDAA_2025Submitted manuscript, 1.43 MBLicence: Unspecified

CUGS: Intelligent Combat Unmanned Ground Systems (iCUGS)
Gallant, J. (Promotor), De cubber, G. (Promotor), Neyt, X. (Promotor), Le Flécher, E. (Promotor), Maroulis, E. (Researcher), Boeckx, K. (Researcher), Vanvolsem, P. (Researcher), Ndindabahizi, I. (Researcher) & Miranda Vicario, A. (Researcher)
1/04/23 → 1/10/26
Project: Research › EDA

Interpreting Tactical Decision-Making in Transformer-Based Agents
Boeckx, K. (Speaker) & Neyt, X. (Co-author)
25 Nov 2025
Activity: Talk or presentation › Oral scientific presentation

Cite this

@inproceedings{42a912bb56c44bdf934c1674e97cfe25,

title = "Interpreting Tactical Decision-making in Transformer-based Agents",

abstract = "The development of autonomous agents capable of sophisticated strategic decision-making in complex environments is a central goal of artificial intelligence. This paper proposes a framework for discovering and interpreting strategies in a simulated grid-world battlefield environment. We leverage the AlphaZero algorithm, a powerful reinforcement learning method that combines Monte Carlo Tree Search (MCTS) with deep neural networks, to train agents. Crucially, the neural network component incorporates a Transformer architecture. The primary contribution of this work lies in the proposed methodology to utilize the self-attention mechanisms within the Transformer to gain insights into the agent{\textquoteright}s decision-making process, specifically by visualizing which parts of the battlefield the network pays attention to when selecting an action. This ",

author = "Koen Boeckx and Xavier Neyt",

year = "2025",

month = nov,

day = "25",

language = "English",

pages = "51--55",

editor = "Yurish, \{Sergey Y.\}",

booktitle = "Proceedings of the 1st International Conference on Big Data \& Applications",

publisher = "International Frequency Sensor Association (IFSA) Publishing",

note = "1st International Conference on Big Data Analytics and Applications, BDAA 2025 ; Conference date: 25-11-2025 Through 27-11-2025",

url = "https://bdaa-conference.com/",

}

Boeckx, K & Neyt, X 2025, Interpreting Tactical Decision-making in Transformer-based Agents. in SY Yurish (ed.), Proceedings of the 1st International Conference on Big Data & Applications: BDAA 2025., 18, International Frequency Sensor Association (IFSA) Publishing, 2025, pp. 51-55, 1st International Conference on Big Data Analytics and Applications, Innsbruck, Austria, 25/11/25.

Interpreting Tactical Decision-making in Transformer-based Agents. / Boeckx, Koen ; Neyt, Xavier.
Proceedings of the 1st International Conference on Big Data & Applications: BDAA 2025. ed. / Sergey Y. Yurish. 2025: International Frequency Sensor Association (IFSA) Publishing, 2025. p. 51-55 18.

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

TY - GEN

T1 - Interpreting Tactical Decision-making in Transformer-based Agents

AU - Boeckx, Koen

AU - Neyt, Xavier

N1 - Conference code: 1

PY - 2025/11/25

Y1 - 2025/11/25

N2 - The development of autonomous agents capable of sophisticated strategic decision-making in complex environments is a central goal of artificial intelligence. This paper proposes a framework for discovering and interpreting strategies in a simulated grid-world battlefield environment. We leverage the AlphaZero algorithm, a powerful reinforcement learning method that combines Monte Carlo Tree Search (MCTS) with deep neural networks, to train agents. Crucially, the neural network component incorporates a Transformer architecture. The primary contribution of this work lies in the proposed methodology to utilize the self-attention mechanisms within the Transformer to gain insights into the agent’s decision-making process, specifically by visualizing which parts of the battlefield the network pays attention to when selecting an action. This

AB - The development of autonomous agents capable of sophisticated strategic decision-making in complex environments is a central goal of artificial intelligence. This paper proposes a framework for discovering and interpreting strategies in a simulated grid-world battlefield environment. We leverage the AlphaZero algorithm, a powerful reinforcement learning method that combines Monte Carlo Tree Search (MCTS) with deep neural networks, to train agents. Crucially, the neural network component incorporates a Transformer architecture. The primary contribution of this work lies in the proposed methodology to utilize the self-attention mechanisms within the Transformer to gain insights into the agent’s decision-making process, specifically by visualizing which parts of the battlefield the network pays attention to when selecting an action. This

M3 - Conference contribution

SP - 51

EP - 55

BT - Proceedings of the 1st International Conference on Big Data & Applications

A2 - Yurish, Sergey Y.

PB - International Frequency Sensor Association (IFSA) Publishing

CY - 2025

T2 - 1st International Conference on Big Data Analytics and Applications

Y2 - 25 November 2025 through 27 November 2025

ER -

Interpreting Tactical Decision-making in Transformer-based Agents

Abstract

Conference

Access to Document

Fingerprint

Projects

CUGS: Intelligent Combat Unmanned Ground Systems (iCUGS)

Activities

Interpreting Tactical Decision-Making in Transformer-Based Agents

Cite this