Announging two new journal publications:

• “Health State Prediction with Reinforcement Learning for Predictive Maintenance,” in the journal Frontiers in Artificial Intelligence, 12 January 2026, Volume 8 – 2025, by Anastasis Aglogallos, Alexandros Bousdekis, Stefanos Kontos, Gregoris Mentzas.
  • Traditional predictive maintenance (PdM) models often struggle with the need for extensive labeled data and adapting to evolving operational conditions. Our work addresses this problem by formulating the PdM problem as a Markov Decision Process (MDP) and leveraging advanced model-free Reinforcement Learning (RL) algorithms. We demonstrate how RL agents can learn optimal health state prediction policies by interacting with the environment and validate our approach on CNC machine tool wear data. Our research underscores the immense potential of RL to revolutionize maintenance strategies by naturally capturing the sequential and uncertain dynamics of equipment degradation.
  • Read the full article: https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1720140/full
• “Generative AI for Autonomous Data Analytics”, Intelligent Systems with Applications, Volume 29, March 2026, by Mattheos Fikardos, Katerina Lepenioti, Alexandros Bousdekis, Dimitris Apostolou, Gregoris Mentzas
  • As large language models and generative AI reshape the frontiers of automation, there’s a growing need to rethink how data analytics platforms interact with users and analytical logic. In this work, we propose a framework that embeds generative AI directly into analytics workflows, enabling smarter, more autonomous collaboration between analysts, AI agents, and analytics infrastructure. We reengineer traditional analytics platform APIs into a set of tools usable by AI agents, enabling them to autonomously orchestrate and execute analytical tasks. Our approach includes a chat-based interface where analysts can express analytical goals in natural language, seamlessly triggering workflows without deep technical expertise. The framework facilitates effective coordination among human analysts, AI agents, and backend analytics systems — enhancing both accessibility and productivity. Instantiated with open-source models and evaluated across multiple scenarios, our system achieved a ~7.2 % improvement in overall analytical performance compared to baselines, while complementary user-study data revealed that the chat-based interface led to higher task efficiency and stronger user preference versus traditional form-based analytics.
  • Read the full paper: https://www.sciencedirect.com/science/article/pii/S2667305326000013