Abstract:

Reactive (Program) Synthesis is an area of Formal Methods that studies how to automatically synthesize interactive programs (reactive systems) from a human-readable specification, typically expressed in temporal logic. In this talk, we will discuss its use in Autonomous AI Systems to equip them with strategic reasoning capabilities, as well as frame such capabilities within formally specified bounds that provide guardrails for their self-deliberated behavior. We will show that Reactive Synthesis is related to certain forms of AI Planning in partially controllable environments and to MDP solving. Technically, we will focus on specifications expressed in Linear Temporal Logic (LTL) and, in particular, its finite-trace variants such as LTL_f. A key advantage of these finite-trace variants is their simplicity, due to their reducibility to equivalent regular automata, which can be easily determinized and transformed into two-player games on graphs. This simplicity leads to an unprecedented computational effectiveness and scalability of synthesis in LTL_f compared to LTL. Finally, we will also show how to lift this finite-trace "technology" to infinite traces by leveraging Manna and Pnueli’s safety-progress hierarchy for temporal properties, which is ultimately based on a finite-trace core.

Abstract:

Relational data capture rich, structured dependencies that traditional learning methods struggle to exploit. Relational neural networks aim to model such data by integrating the relational reasoning principles underlying graph neural networks and graph transformers. This talk explores the theoretical foundations of these models and how their learning behaviour interacts with data heterogeneity, temporal dynamics, and schema structure. We question how to characterise what relational architectures can represent and how they learn. Finally, we outline prospects for principled foundation models that unify representation, reasoning, and transfer across diverse relational domains.

Abstract:

Reasoning is fundamental to the foundations of information and knowledge systems, enabling these systems to derive reliable conclusions from complex data. Research in FoIKS focuses on the development of logical, non-monotonic, and probabilistic reasoning methods for addressing uncertainty and incomplete information—capabilities that are essential for intelligent agents and robust knowledge-based systems. Effective reasoning empowers information systems to adapt, update, and validate knowledge as new data becomes available.

For LLM-based agents, reasoning is indispensable, as it facilitates planning, decision-making, problem-solving, and adaptation to dynamic environments in a manner that is both human-like and interpretable. Robust reasoning enables these agents to operate autonomously, tackle complex tasks through reflection, and provide justifications for their actions. These abilities are critical for tasks that require multi-step logic, effective tool use, and proficient interaction with humans or other agents.

This talk will examine whether these two perspectives on reasoning share common ground, highlighting recent research in LLM-related reasoning and discussing realistic expectations for future developments.

Abstract:

Ontology-based data access (OBDA) is a popular paradigm for data integration. In this paradigm, multiple data sources are integrated through mappings to an ontology. These data sources can be heterogeneous (databases, language models, knowledge graph embedding models, etc) and the information retrieved from them may have different levels of trust. So, when answering queries, one may well be interested in not only obtaining the answers but also knowing why/how these answers were obtained. Provenance here refers to the origin of a query result, in particular, which tuples in a database or which data sources were relevant for answering a query. This corresponds to a notion of explanation explored in the context of databases using commutative semirings. In this talk, I will present work on semiring provenance in the context of ontology-based data access. The focus will be on ontologies expressed in DL-Lite and solutions to determine the provenance of a query within the OBDA setting.