Alessandro Antonucci

2026

1. Journal

   Deep Reinforcement Learning in Wargaming

   Giacomo Del Rio, Oleg Szehr, Marco Pierbattista, Alessandro Antonucci, Michael Rüegsegger, and 1 more author

   Journal of Aerospace Information Systems, Jan 2026

   Accepted

   Abs Bib PDF

   We investigate the potential of deep reinforcement learning (RL) for the development of autonomous wargaming agents. We discuss the relevant characteristics of wargaming environments for the design of learning systems, the choice of learning framework and algorithms. While deep RL has been demonstrated to achieve superhuman levels in various games, we argue that these findings can be only partially transferred to practical wargaming. This is due to real-world limitations such as the availability of financial and data resources, but also architectural system requirements that might rarely be satisfied in the wargaming area. The high degree of realism of modern warfare simulation environments is often accompanied with a system latency that entails impractical training times. For an empirical analysis, we adapt various deep RL techniques to the popular Command: Modern Operations simulation environment providing a proof-of-concept for deep RL training applications in this environment.

   @article{delrio2026,
     author = {Rio, Giacomo Del and Szehr, Oleg and Pierbattista, Marco and Antonucci, Alessandro and Rüegsegger, Michael and Sommer, Matthias},
     journal = {Journal of Aerospace Information Systems},
     title = {Deep Reinforcement Learning in Wargaming},
     publisher = {AIAA},
     year = {2026},
     month = jan,
     note = {Accepted},
     bibtex_show = true,
     selected = true
   }

2025

1. Proceedings

   Causal Understanding by LLMs: The Role of Uncertainty

   Oscar Lithgow-Serrano, Vani Kanjirangat, and Alessandro Antonucci

   In Proceedings of the Second Workshop on Uncertainty-Aware NLP (UncertaiNLP 2025), Nov 2025

   Abs Bib PDF

   Recent papers show LLMs achieve near-random accuracy in causal relation classification, raising questions about whether such failures arise from limited pretraining exposure or deeper representational gaps. We investigate this under uncertainty-based evaluation, testing whether pretraining exposure to causal examples improves causal understanding using >18K PubMed sentences -half from The Pile corpus, half post-2024- across seven models (Pythia-1.4B/7B/12B, GPT-J-6B, Dolly-7B/12B, Qwen-7B). We analyze model behavior through: (i) causal classification, where the model identifies causal relationships in text, and (ii) verbatim memorization probing, where we assess whether the model prefers previously seen causal statements over their paraphrases. Models perform four-way classification (direct/conditional/correlational/no-relationship) and select between originals and their generated paraphrases. Results show almost identical accuracy on seen/unseen sentences (p>0.05), no memorization bias (24.8% original selection), output distribution over the possible options almost flat - with entropic values near the maximum (1.35/1.39), confirming random guessing. Instruction-tuned models show severe miscalibration (Qwen: 95% confidence, 32.8% accuracy, ECE=0.49). Conditional relations induce highest entropy (+11% vs direct). These findings suggest that failures in causal understanding arise from the lack of structured causal representation, rather than insufficient exposure to causal examples during pretraining.

   @inproceedings{lithgow2025,
     author = {Lithgow-Serrano, Oscar and Kanjirangat, Vani and Antonucci, Alessandro},
     booktitle = {Proceedings of the Second Workshop on Uncertainty-Aware NLP (UncertaiNLP 2025)},
     title = {Causal Understanding by {LLM}s: The Role of Uncertainty},
     year = {2025},
     month = nov,
     pages = {208--228},
     publisher = {Association for Computational Linguistics},
     editor = {Eikema, Bryan and V{\'a}zquez, Ra{\'u}l and Berant, Jonathan and de Marneffe, Marie-Catherine and Plank, Barbara and Shelmanov, Artem and Swayamdipta, Sawabha and Tiedemann, J{\"org} and Zerva, Chrysoula and Aziz, Wilker},
     address = {Suzhou, China},
     url = {https://aclanthology.org/2025.uncertainlp-main.19},
     bibtex_show = true,
     doi = {10.18653/v1/2025.uncertainlp-main.19},
     selected = true
   }

2. Proceedings

   Understanding Action Effects through Instrumental Empowerment in Multi-Agent Reinforcement Learning

   Ardian Selmonaj, Miroslav Strupl, Oleg Szehr, and Alessandro Antonucci

   In Proceedings of the 28th European Conference on Artificial Intelligence (ECAI 2025), Oct 2025

   Abs arXiv Bib HTML PDF

   To reliably deploy Multi-Agent Reinforcement Learning (MARL) systems, it is crucial to understand individual agent behaviors. While prior work typically evaluates overall team performance based on explicit reward signals, it is unclear how to infer agent contributions in the absence of any value feedback. In this work, we investigate whether meaningful insights into agent behaviors can be extracted solely by analyzing the policy distribution. Inspired by the phenomenon that intelligent agents tend to pursue convergent instrumental values, we introduce Intended Cooperation Values (ICVs), a method based on information-theoretic Shapley values for quantifying each agent’s causal influence on their co-players’ instrumental empowerment. Specifically, ICVs measure an agent’s action effect on its teammates’ policies by assessing their decision (un)certainty and preference alignment. By analyzing action effects on policies and value functions across cooperative and competitive MARL tasks, our method identifies which agent behaviors are beneficial to team success, either by fostering deterministic decisions or by preserving flexibility for future action choices, while also revealing the extent to which agents adopt similar or diverse strategies. Our proposed method offers novel insights into cooperation dynamics and enhances explainability in MARL systems.

   @inproceedings{selmonaj2025a,
     author = {Selmonaj, Ardian and Strupl, Miroslav and Szehr, Oleg and Antonucci, Alessandro},
     booktitle = {Proceedings of the 28th European Conference on Artificial Intelligence (ECAI 2025)},
     title = {Understanding Action Effects through Instrumental Empowerment in Multi-Agent Reinforcement Learning},
     year = {2025},
     month = oct,
     publisher = {IOS Press},
     bibtex_show = true,
     doi = {10.3233/FAIA250947},
     selected = true
   }

3. Proceedings

   Counterfactual Inference Using Ordinary Differential Equations to Assess the Effect of Physical Activity on Type 2 Diabetes Onset

   Marta Lenatti, Marco Zaffalon, Alessandro Antonucci, Pierluigi De Paola, Lea Multerer, and 3 more authors

   In Proceedings of the 23rd International Conference on Artificial Intelligence in Medicine (AIME 2025), Jun 2025

   Abs Bib HTML

   This study presents a counterfactual inference approach for ordinary differential equations (ODEs) models. The methodology is applied to a model describing blood glucose regulation to assess the effects of personalized physical activity plans on type 2 diabetes progression. The analysis demonstrates how responses to physical activity vary among individuals based on beta-cells mass and insulin sensitivity, resulting in diverse sets of counterfactual physical activity plans. This approach can eventually help identify individuals who might benefit from early diabetes prevention measures via personalized physical activity plans.

   @inproceedings{lenatti2025a,
     author = {Lenatti, Marta and Zaffalon, Marco and Antonucci, Alessandro and Paola, Pierluigi De and Multerer, Lea and Mongelli, Maurizio and Paglialonga, Alessia and Azzimonti, Laura},
     booktitle = {Proceedings of the 23rd International Conference on Artificial Intelligence in Medicine (AIME 2025)},
     title = {Counterfactual Inference Using Ordinary Differential Equations to Assess the Effect of Physical Activity on Type 2 Diabetes Onset},
     year = {2025},
     editor = {Bellazzi, Riccardo and Herrero, Jos{\'e} Manuel Juarez and Sacchi, Lucia and Zupan, Bla{\v{z}}},
     month = jun,
     pages = {212--221},
     publisher = {Springer Nature Switzerland},
     bibtex_show = true,
     selected = true
   }