In this paper, we present a reproducible pipeline to benchmark the trade-off between carbon emissions and recommendation performance across 14 algorithms and three publicly available datasets. In particular, we contribute: (a) a standardized protocol to account for carbon emissions of recommendation algorithms; (b) an empirical quantification of the carbon cost of hyperparameter tuning, and (c) an evaluation of data-reduction strategies as a low-cost approach to reduce emissions while improving certain non-accuracy metrics. Unlike previous literature, which mainly focused on the trade-off between performance and emissions, our benchmark reveals the cost of hyperparameter tuning. It examines the impact of data reduction techniques on the path toward sustainability-aware recommender systems. Our results show that simpler algorithms often deliver competitive accuracy at significantly lower emissions, and that exhaustive tuning can dramatically increase carbon costs with limited accuracy gains. Generally speaking, this study aims to discuss the challenges of energy consumption in recommender systems and to develop a new generation of algorithms that prioritize sustainability. All code and experiment traces are publicly released for reproducibility on Github.1

High-Performance Computing (HPC) systems increasingly require intelligent, scalable anomaly detection to ensure operational reliability. However, conventional centralized approaches often struggle with data privacy constraints, poor generalization across heterogeneous nodes, and limited scalability. This study presents the first real-world application of federated transfer learning (FTL) for anomaly detection in a production-grade Tier-0 supercomputer. By combining federated learning with transfer learning, the proposed framework enables decentralized model training and personalized adaptation to unseen nodes, without accessing raw data. We validate the approach using two large-scale telemetry datasets collected from 100 nodes of the Marconi100 supercomputer, evaluating its effectiveness across supervised, semi-supervised, and unsupervised learning paradigms. Results show that FTL consistently improves anomaly detection performance on nodes that did not participate in federated training, with F1-score gains reaching up to 0.50. These improvements demonstrate the framework's ability to generalize across non-identically distributed data and maintain detection accuracy under real-world conditions. This work establishes FTL as a scalable, privacy-preserving solution for fault detection in HPC environments. Its practical deployment on production hardware confirms its readiness for real-time monitoring applications in large-scale, heterogeneous computing systems.

This paper introduces a preliminary implementation of the Feature Mapping module, a core component of the novel approach that exploits swarm intelligence algorithms to generate dance choreography by virtual agents. We delegate the mapping process from music to simulation parameters to a neural network, casting its development as an optimization problem. Additionally, we propose an initial qualitative evaluation apparatus in the form of a questionnaire to assess the performances generated. We leverage preliminary human evaluation results to assess the automatic evaluation system employed, specifically the fitness functions and hyperparameters used. This work represents a first step towards the realization of a human-in-the-loop framework for the generation of artistic performances. Moreover, this leads to the realization of eXtended Reality application scenarios.

The concept of sustainability, as outlined by the United Nations’ Sustainable Development Goals (SDGs), refers to the ability to meet the needs of the present without compromising the ability of future generations to meet their own needs. This vision is addressed by combining goals concerning the environmental, social, and economic spheres. In this context, Recommender Systems (RS) have emerged as tools that can foster these principles by nudging responsible user behavior and promoting sustainable decision-making. However, the interplay between RS and sustainability is inherently complex since it can be analyzed from two different perspectives: (i) RS for Sustainability, which focuses on how recommendation algorithms can support the achievement of SDGs, and (ii) Sustainability of RS, which focuses on developing recommendation models that inherently adhere to sustainability principles. While the integration of both these perspectives is beneficial and crucial, unfortunately, the current literature has addressed these aspects independently. Accordingly, in this survey, we first provide a comprehensive review of the existing literature on RS that either promotes sustainable behaviors aligned with the SDGs or embeds sustainability principles into their algorithmic design. Next, we identify current gaps and propose key research directions toward an integrated, holistic approach that concurrently addresses both aspects to advance the development of sustainable RS.

High-Performance Computing (HPC) systems are becoming increasingly vulnerable to anomalies as their scale and complexity grow. In this work, we propose a federated learning (FL) framework that integrates Long Short-Term Memory (LSTM) autoencoders for time series anomaly detection, allowing decentralized model training without sharing raw data. Using real telemetry from the Marconi100 Tier-0 supercomputer, our approach improves the average F1-score from 0.388 to 0.867 (+123%) and the AUC from 0.334 to 0.808 (+142%). It also cuts the training data requirement by a factor of 15, reducing the collection period from 4.5 months to just 1.25 weeks. These improvements are consistent across unsupervised, semi-supervised, and supervised settings, and significance testing with the Wilcoxon signed-rank test confirms they are statistically robust (p < 0.01). To our knowledge, this is the first comprehensive evaluation of FL-based LSTM autoencoders for anomaly detection in real HPC environments.

The literature on AI fairness has grown rapidly, proposing a large number of bias mitigation techniques that are commonly organized into pre-, in-, and post-processing methods. This pipeline-centric view offers an operational, lifecycle-based perspective on where mitigation can be applied. In deployment settings, however, practitioners also face an additional question: whether a mitigation family is applicable given the resources and access rights available in a concrete system. In this survey, we use resources broadly to denote data access/control, training capability, and deployment-time interface/decision control. Accordingly, we introduce a resource-aware taxonomy that complements existing taxonomies by classifying AI bias mitigation methods according to the conditions that make them practically implementable. We use this taxonomy to structure and reinterpret existing literature on the topic, highlighting which mitigation families remain feasible under resource constraints.

In the era of responsible and sustainable AI, information retrieval and recommender systems must expand their scope beyond traditional accuracy metrics to incorporate environmental sustainability. However, this research line is severely limited by the lack of item-level environmental impact data in standard benchmarks. This paper introduces Eco-Amazon, a novel resource designed to bridge this gap. Our resource consists of an enriched version of three widely used Amazon datasets (i.e., Home, Clothing, and Electronics) augmented with Product Carbon Footprint (PCF) metadata. CO2e emission scores were generated using a zero-shot framework that leverages Large Language Models (LLMs) to estimate item-level PCF based on product attributes. Our contribution is three-fold: (i) the release of the Eco-Amazon datasets, enriching item metadata with PCF signals; (ii) the LLM-based PCF estimation script, which allows researchers to enrich any product catalogue and reproduce our results; (iii) a use case demonstrating how PCF estimates can be exploited to promote more sustainable products. By providing these environmental signals, Eco-Amazon enables the community to develop, benchmark, and evaluate the next generation of sustainable retrieval and recommendation models. Our resource is available at https://doi.org/10.5281/zenodo.18549130, while our source code is available at: http://github.com/giuspillo/EcoAmazon/.

The environmental impact of e-commerce continues to grow, driven by global supply chains and unsustainable consumption patterns. In this context, accurately estimating product carbon footprint (PCF) is a fundamental step to making decisions that are more environmentally conscious. However, this process is complex to implement due to the lack of comprehensive information about the carbon footprint of items and production processes. Accordingly, this study introduces a novel methodology that leverages Large Language Models (LLMs) to estimate the life cycle carbon footprint ( CO_2 ) of commercial products using unstructured textual data, such as product descriptions and metadata. The approach enables the automatic augmentation of product datasets with environmental indicators. To demonstrate the practical relevance of this framework, we integrate the CO_2 information into a recommender system, enabling the generation of personalized yet more environmentally sustainable suggestions. Experimental results on an Amazon Electronics dataset confirm the effectiveness of LLMs in approximating emission values, offering a reliable strategy to support both research and policy efforts targeting Goal 12 of the UN Sustainable Development Goals: Responsible Consumption and Production.

This paper explores the potential of generative artificial intelligence (GenAI) in the artistic domain of robotic dance, by focusing on Human-AI co-creation. Specifically, we evaluate the capabilities of three state-of-the-art Large Language Models (LLMs) in generating robotic dance choreographies in collaboration with human artists. We design and test three prompting techniques tailored for robotic dance creation, in which we progressively introduce human expertise of choreographers in the form of examples and natural language feedback through an iterative process. In this way, we analyzed the dynamics of human-AI co-creation across different LLMs on varying of the parameters of the models and of different design of the prompts. The experimental analysis is conducted through a quantitative and qualitative evaluation, with the help of a human audience composed by 210 participants, and using a state-of-the-art evaluation scheme. The results revealed that each model shows different strengths and limitations. Moreover, when enriched by human intervention, the output generated from LLMs generally improve its artistic impact. These results highlight the potential of different LLM architectures in the artistic domain and in creative tasks. Ultimately, starting from the artistic domain of dance, this work contributes a modular methodology to enhance the intersection of human creativity and AI-driven creations, to develop frameworks for human-AI collaboration.

Recent developments in artificial intelligence (AI) foster creativity in all its forms. In particular, with the advent of extended reality (XR), we have witnessed an explosion of once unimaginable artistic creations and products, one above all the Metaverse. Dance and the process of creating dance choreography can benefit from the use of these new technologies, but so far little has been done in this regard. This paper provides an overview of the existing literature on the use of AI tools in combination with XR technology within the artistic context of dance. Furthermore, we propose new application scenarios that involve human interactions mediated by XR technologies. In particular, we focus on dance choreography performed by artificial agents driven by Swarm Intelligence mechanisms, and we discuss the perspectives and potential implications for both artificial and human creativity through the integration with XR.

Environmental sustainability of AI, or Green AI, is a topic that is getting more and more crucial in the last few years. However, AI systems continue to improve at the cost of huge resources, neglecting the environmental impact in terms of CO_2 emissions from computations. In this context, Recommender Systems (RS) are no exception, and the current literature in the field pays little attention to the concept of Green AI. In this paper, we propose a tool that aims at estimating the CO_2 emitted by a recommendation model. Our contributions are twofold: first, we built a regression dataset that can be used to feed a regression model aiming at estimating the emissions or RS models; this dataset can be easily expanded, so it can be considered a relevant resource for the whole community. Second, we compared several state-of-the-art regression models to assess which performs the best and in which settings. Results show that Random Forest is the best performing model to effectively estimate the CO_2 emissions produced by recommendation models.

With the growing focus on Green AI, there is an urgent need for algorithms that are designed to minimize their environmental impact while maintaining satisfying performance. In this paper, we introduce a novel early stopping strategy that considers carbon footprint data while training a recommendation algorithm. In particular, during the training phase, our criterion epoch-by-epoch analyzes the improvement in terms of predictive accuracy and compares it to the increase in carbon emissions. Then, we analyze the trade-off between the scores, and when the accuracy improves at a rate that is not favorable, the training is stopped. In the experimental evaluation, we showed that our strategy could significantly reduce the carbon footprint of several state-ofthe-art recommendation models, with a limited decrease in accuracy and fairness. While more work is needed to automatically balance the trade-off between accuracy and emissions, this paper sheds light on the need for more sustainable recommendation models and takes a significant step toward designing green training strategies.

Traditional university lessons do not provide students with the opportunity to put theoretical concepts into practice. Project-Based Learning is designed to involve students through the proposition of real-word problems in the form of a project. The main objective of this work is to improve several aspects of the student's learning experience (e.g., their motivation and interest) through practical experience during a master degree course in Artificial Intelligence. We propose an application of Project-Based Learning through the use of a game-based competition. The experience is designed as an activity that takes place in parallel with respect to the usual lessons. Then, we assessed the significance and the impact of this approach, from the educational point of view, through questionnaires proposed to the students involved. The results of a 3-year study involving more than 200 students are positive. Students reacted favorably to the experience: they think this experience improved their knowledge of AI, their motivation, and their skills. The competitive aspect is considered beneficial from multiple perspectives. Finally, the design of the experience seems to be robust and remains effective in a setting of remote lectures.

As recommendation algorithms become increasingly sophisticated and pervasive, their energy consumption and associated carbon emissions are rising significantly. To address this growing environmental concern, this work investigates the path toward ‘green recommender systems’ by examining how data reduction techniques can impact on algorithm performance and carbon footprint. We specifically investigated whether and how a reduction of the training data impacts the performance of several representative recommendation algorithms. To obtain a fair comparison, all the algorithms were run based on the implementations available in a popular recommendation library, i.e., RecBole, and by using the same experimental settings. Specifically, we employed distinct data reduction strategies: (a) random sampling of either users or item ratings; (b) reducing the overall dataset size; (c) filtering out more recent user ratings. Results indicate that data reduction can be a promising strategy to make recommender systems more environmentally sustainable with a relevant reduction in carbon emissions at the cost of a smaller reduction in predictive accuracy -43.38% of emissions for LightGCN algorithm and only 3.72% loss in accuracy in a book recommendation scenario). Moreover, training recommender systems with less data makes the suggestions less prone to popularity bias. Overall, this study contributes to the ongoing challenge of developing recommendation algorithms that meet the principles of Sustainable Development Goals, by proposing the adoption of more sustainable practices in the field.

Fairness is a critical concern in Machine Learning, impacting its applications across domains. Existing fairness analyses often rely on complex mathematics, lacking of intuitive understanding. In this study, we introduce \emph{GEOFFair}, a Geometric Framework for Fairness. It represents Machine Learning elements as vectors and sets, offering a more intuitive understanding of fairness related concepts. GEOFFair visualizes fairness mitigation techniques as vector projections, it provides a solid base to investigate the bias injection, aiding in constructing proofs, and it enables the study of fairness properties by means of geometric considerations. The main contribution of the work is to highlight GEOFFair's effectiveness in fairness studies, demonstrating that solely maximizing accuracy based on observed labels may not always be optimal for fairness.

This work investigates the path toward green recommender systems by examining the impact of data reduction on both model performance and carbon footprint. In the pursuit of developing energy-efficient recommender systems, we investigated whether and how reducing the training data impacts the performances of several representative recommendation models. In order to obtain a fair comparison, all the models were run based on the implementations available in a popular recommendation library, i.e., RecBole, and used the same experimental settings. Results indicate that: (a) data reduction can be a promising strategy to make recommender systems more sustainable, at the cost of a lower accuracy; (b) training recommender systems with less data makes the suggestions more diverse and less biased. Overall, this study contributes to the ongoing discourse on the development of recommendation models that meet the principles of SDGs, laying the groundwork for the adoption of more sustainable practices in the field.

This paper focuses on the potential of Generative Artificial Intelligence (AI), particularly Large Language Models (LLMs), in the still unexplored domain of robotic dance creation. In particular, we assess whether a LLM (GPT-3.5 turbo) can create robotic dance choreographies, and we investigate if the feedback provided by human creators can improve the quality of the output. To this end, we design three prompt engineering techniques for robotic dance creation. In the prompts, we gradually introduce human knowledge through examples and feedback in natural language in order to explore the dynamics of human-AI co-creation. The experimental analysis shows that the capabilities of the LLM can be improved through human collaboration, by producing choreographies with a major artistic impact on the evaluation audience. The findings offer valuable insights into the interplay between human creativity and AI generative models, paving the way for enhanced collaborative frameworks in creative domains.

Recent advancements in Artificial Intelligence in Education (AIEd) have revolutionized educational practices using machine learning to extract insights from students' activities and behaviours. Performance prediction, a key domain within AIEd, aims to enhance student achievement levels and address sustainable development goals related to education, health, gender equality, and economic growth. However, the potential of AIEd to contribute to these goals is hindered by the lack of attention to fairness in prediction algorithms, leading to educational inequality. To address this gap, we introduce FAiRDAS a general framework that models long-term fairness as an abstract dynamic system. Our approach, illustrated through a case study in AIEd with real data, offers a customizable solution to promote long-term fairness while promoting the stability of mitigation actions over time.