Publications

You can also find my articles on my Google Scholar profile.

Conference Papers

On Planning Through LLMs

Published in 35th. International Conference on Automated Planning and Scheduling (ICAPS), 2025

This paper reviews recent research on using Large Language Models (LLMs) for automated planning tasks. In these studies, LLMs are typically given a planning domain along with an initial state and a goal, and asked to generate a sequence of actions (a plan) that solves the problem. While the setup is similar across studies, they vary in the models used, the information provided, whether symbolic planners are used, and how results are evaluated. We summarize the main research trends, notable findings, and current challenges in evaluating LLMs’ planning abilities. link paper

Recommended citation: On Planning Through LLMs, M. Chiari, L. Putelli, N. Rossetti, I. Serina, AE. Gerevini - Proceedings of the 35th. International Conference on Automated Planning and Scheduling (ICAPS), 2025
Download Paper

Integrating Classical Planners with GPT-Based Planning Policies

Published in 33th International Conference of the Italian Association for Artificial Intelligence (AIxIA), 2024

This paper introduces a neuro-symbolic approach to improve the reliability of PlanGPT, a gpt-2 model trained from scratch for automated planning. While PlanGPT effectively learns general planning policies, it can produces incomplete or invalid plans that violate action preconditions or only partially achieve goals. To address this, we integrate PlanGPT with the symbolic planner LPG. After PlanGPT generates a candidate plan, a validator checks its validity. If the plan is flawed, LPG repairs or completes it, ensuring a correct solution. Our results show significant improvements in the performance of both PlanGPT and LPG, highlighting the effectiveness of combining learning methods with traditional planning techniques. link paper

Recommended citation: Integrating Classical Planners with GPT-Based Planning Policies, M. Tummolo, N. Rossetti, AE. Gerevini, M. Olivato, L. Putelli, I. Serina - Proceedings of the 33th International Conference of the Italian Association for Artificial Intelligence (AIxIA), 2024
Download Paper

Enhancing GPT-based Planning Policies by Model-based Plan Validation

Published in 18th International Conference of Neural-Symbolic Learning and Reasoning (NeSy), 2024

This paper builds on PlanGPT, a GPT-2 model tailored for automated planning tasks, which, despite strong performance, can produce invalid plans that violate preconditions or fail to meet goal fluents. To improve reliability, we propose an enhanced version of PlanGPT that incorporates a plan validator into the token generation process. This validator filters out invalid plan sequences as they are being generated, resulting in more accurate plans. The method is tested across multiple planning domains, showing improved robustness and effectiveness. link paper

Recommended citation: Enhancing GPT-based Planning Policies by Model-based Plan Validation, N. Rossetti, M. Tummolo, AE. Gerevini, M. Olivato, L. Putelli, I. Serina - Proceedings of the 18th International Conference of Neural-Symbolic Learning and Reasoning (NeSy), 2024
Download Paper

Learning general policies for planning through GPT models

Published in 34th International Conference on Automated Planning and Scheduling (ICAPS), 2024

This paper introduces PlanGPT, a GPT-based model customized for classical planning. PlanGPT is trained from scratch on a dataset of solved planning instances to learn a general policy. Once trained, it can generate solution plans for new problems within the same domain. We enhance the model performance by incorporating automated planning knowledge into the training process. We evaluate PlanGPT across several planning domains and compare its performance to other deep learning techniques for generalized planning, demonstrating the effectiveness of our approach. link paper

Recommended citation: Learning general policies for planning through GPT models, N. Rossetti, M. Tummolo, AE. Gerevini, L. Putelli, I. Serina, M. Chiari, M. Olivato - Proceedings of the 34th International Conference on Automated Planning and Scheduling (ICAPS), 2024
Download Paper | Download Slides

Recurrent Neural Networks for Daily Estimation of COVID-19 Prognosis with Uncertainty Handling

Published in 27th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems (KES), 2023

This paper introduces an RNN-based system for daily prognosis estimation of COVID-19 patients. Unlike previous models trained on short-term datasets, this system is built using patient data collected over an extended period, capturing the evolving nature of the pandemic, including the effects of new variants, treatments, and vaccines. Additionally, it incorporates an uncertainty-aware mechanism that discards predictions lacking sufficient confidence, reducing the risk of misleading results. paper link

Recommended citation: Recurrent Neural Networks for Daily Estimation of COVID-19 Prognosis with Uncertainty Handling, N. Rossetti, AE. Gerevini, M. Olivato, L. Putelli, M. Chiari, I. Serina, D. Minisci, E. Foca - Proceedings of the 27th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems (KES), 2023
Download Paper

Machine learning models for predicting short-long length of stay of COVID-19 patients

Published in 26th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems (KES), 2022

This paper presents a machine learning system trained on data from over 6,000 hospitalized patients in northern Italy during the COVID-19 pandemic. The system addresses the challenge of predicting the length of stay for both alive and deceased patients. We introduce a method to handle data from both groups, using information of the outcome of a patient. Additionally, we analyze the most relevant features for predicting the length of stay, providing insights into key factors influencing hospitalization durations. link paper

Recommended citation: Machine learning models for predicting short-long length of stay of COVID-19 patients, M. Olivato, N. Rossetti, AE. Gerevini, M. Chiari, L. Putelli, I. Serina - Proceedings of the 26th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems (KES), 2022
Download Paper

An Application of Recurrent Neural Networks for Estimating the Prognosis of COVID-19 Patients in Northern Italy

Published in Proceedings of the 19th International Conference on Artificial Intelligence in Medicine (AIME), 2021

This paper presents a predictive tool based on Recurrent Neural Networks (RNNs) to estimate the risk of death for hospitalised COVID-19 patients. The goal is to help hospitals manage limited resources—such as ICU beds and ventilators, by identifying high-risk patients in advance. The model uses demographic data, lab test results, and a lung damage severity score from chest X-rays as input features. Trained and tested on data from 2,000 patients in Lombardy, Italy, the system demonstrates strong performance. paper link

Recommended citation: An Application of Recurrent Neural Networks for Estimating the Prognosis of COVID-19 Patients in Northern Italy, M. Chiari, AE. Gerevini, M. Olivato, L. Putelli, N. Rossetti, I. Serina - Proceedings of the 19th International Conference on Artificial Intelligence in Medicine (AIME), 2021

Nicholas Rossetti