Activities
Academic Seminars / Invited Talks
Reasoning at the Right Level: Abstraction and Synthesis of High-Level Programs
University of Liverpool (Liverpool, UK) · June 2026
[]
[slides]
High-level programming languages such as Golog can be seen as a middle layer between classical planning and conventional programming: the modeler sketches intended behaviour, leaving open choices to the reasoner. When the domain is nondeterministic and the environment adversarial, executing such a program becomes a synthesis problem — one that admits a clean game-theoretic characterization with unexpectedly low complexity, thanks to the procedural structure of the program. The same nondeterministic setting offers a natural account of abstraction: a high-level theory can stand in for an infinite family of concrete problem instances, and a strategy synthesized at the abstract level refines into a correct solution for all of them. The talk will explore these two threads, and ask whether abstraction can make strategic reasoning not only general, but practical.
Selected Topics in Symbolic and Neuro-Symbolic AI - Research Seminar
Sun Yat-sen University (Guangzhou, Guangdong, China) · January 2026
[]
This seminar presents an overview of selected research topics in symbolic and neuro-symbolic artificial intelligence. I will discuss recent work on reasoning and planning in dynamic and nondeterministic domains, focusing on logical formalisms such as the Situation Calculus and Golog for strategy synthesis and controller construction with formal guarantees. I will also introduce a complementary line of research on neuro-symbolic predictive process monitoring, where neural sequence models are integrated with temporal logic constraints to ensure logically consistent predictions. The talk aims to provide a coherent overview of ongoing research efforts and future directions.
From Abstract Plans to Concrete Strategies: Synthesizing Controllers in Nondeterministic Domains via Situation Calculus and Golog
City St George's, University of London (London, UK) · April 2025
[]
[slides]
Logical frameworks for planning under uncertainty often rely on abstracting complex domains and reasoning over temporally extended goals. The situation calculus and Golog are well-established formalisms for modeling dynamic systems and specifying high-level agent behavior. This talk presents two complementary approaches to strategy synthesis in nondeterministic domains. The first introduces a notion of temporal abstraction, where a family of concrete planning problems is captured by a single propositional nondeterministic model. Strategies are synthesized at the abstract level and can be provably refined to work across all concrete instances via a sound refinement mapping. The second approach explores the use of Golog programs as procedural specifications in FOND domains. By compiling such programs into symbolic automata, we enable the application of automata-theoretic methods to construct robust controllers that realize the specified behaviors under adversarial environments.
Service
I served as a PC member for the following conferences:
- AAAI 2026
- Workshop on SKILLED-LLMs 2026
I served as a reviewer for the following conferences:
- Workshop on ANNPR 2026
- ECAI 2025
- KR 2025 (KR&R in P&S track)