9:30 Registration
10:00-11:15 Gianluca Aguzzi, Developing Autonomous Agents in Practice with LangChain > This tutorial offers a hands-on introduction to building intelligent agents using large language models (LLMs) and the LangChain library. We will begin by exploring different ways to use large language models — whether through API access or by running them locally — and then present LangChain as a general framework to integrate and orchestrate these capabilities. Along the way, we’ll cover how to enhance LMs with external tools and reasoning strategies using techniques such as prompt engineering, retrieval-augmented generation (RAG), and reasoning and acting (ReAct). These techniques will empower our agents to interact with the real world — browsing the web, generating images, performing calculations, and more.
11:15-12:30 Daniela Briola, Exploiting autonomous agents as "intelligent avatars" into VR environments > Achieving high realism in VR simulations where many virtual autonomous humans (VH) are present is a complex modeling task, and we need to take care of many aspects: in particular, in this talk we focus on how we can improve the autonomy of the virtual humans so that they can really show intelligent behaviours and ability to interact with the environment. Such aspects have been studied for years in the research field of autonomous intelligent agents and multiagent systems, so here we would like to make an analysis of which aspects from those communities can be naturally exploited in VR for making the autonomous virtual humans behave in a more intelligent way, and to identify what is needed in a VR simulation tool to support such improved VHs. At the same time, offering a way to simply perceive and affect the VR environment could offer the MASs a natural new simulation environment, so that they can benefit from a more realistic and "real world - like" environment. We will also have a look at initial activities aiming at realizing such integration, in particular between Unity 3D, JaCaMo and ML agents.
12:30-14:00 Lunch break
14:00-15:15 Giovanni Ciatto and Matteo Magnini, Symbolic Knowledge Extraction and Injection: Theory and Methods > Artificial intelligence has traditionally been approached from two complementary perspectives: symbolic AI, which relies on structured, formal representations of knowledge, and sub-symbolic AI, which leverages data-driven learning methods, such as deep neural networks. While symbolic approaches offer interpretability and logical reasoning, sub-symbolic methods provide flexibility and scalability in processing large datasets. To fully leverage their respective strengths, a growing body of research explores techniques for integrating these paradigms. This talk aims to provide a structured overview of how symbolic and sub-symbolic AI can be connected. We will begin by clarifying the distinction between symbolic and sub-symbolic representations, introducing key symbolic formalisms, and recalling fundamental machine learning concepts. We will then define Symbolic Knowledge Extraction (SKE), which seeks to distil structured knowledge from sub-symbolic models, and Symbolic Knowledge Injection (SKI), which incorporates symbolic knowledge into learning processes. A taxonomy of SKE methods will be presented, followed by an overview of SKI techniques, along with concrete examples of algorithms that instantiate these approaches. Finally, we will discuss applications that benefit from combining symbolic and sub-symbolic methods, as well as potential directions for future research.
15:15-16:30: Giovanni De Gasperis, The DALI 2025 Logic Multi-agent Systems Framework in the era of LLMs > DALI is a Sicstus Prolog extension to develop logic event-based multi-agent systems. The tutorial shows how to develop DALI MASs and integrate them with large language models to acquire factual information from external sources and/or user interactions.
16:30-17:00 Coffee break