Our Research: Core Research Pillars

AILARS research activities are organised around a core number of long-term, stable research pillars. These pillars define shared intellectual directions rather than isolated projects, enabling coherence across institutions, domains, and funding programs.

1  Integrated Neurosymbolic Systems

Architectures that tightly integrate neural perception with symbolic representation, reasoning, and learning, enabling verification and runtime decision-making, communication, and action. Emphasis is placed on unified online–offline pipelines rather than independent, sequential modules.

2  Specification-Guided Learning and Reasoning

Methods that embed real-world specifications—such as logical, physical, regulatory, and safety constraints—directly into learning and inference, reducing reliance on manual labels and post-hoc validation.

3  Human-in-the-Loop Knowledge and Programming

High-level knowledge abstractions—such as domain-specific languages (DSLs) and interaction mechanisms—that enable domain experts to express rules, constraints, and intent without requiring deep machine learning expertise, thereby supporting collaborative human–AI workflows.

4  Automated Specification Mining

Techniques for extracting candidate rules, invariants, and temporal properties from real-world artefacts—such as logs, demonstrations, and system traces—to support the continuous refinement of models, specifications, and runtime monitors.

5  Runtime Safety, Verification, and Provenance

Runtime monitoring, symbolic validation, and provenance tracking mechanisms that support certification, accountability, and regulatory compliance of AI systems.

6  System Interaction for Learning

Integration with feedback guidance between symbolic and neural systems, so multiple forms of learning, e.g. induction, deduction and analogy, can occur in concert.