Logic-guided Agents

Title of the Talk:Logic-guided Agents
Speakers: Divyagna Bavikadi
Host Faculty: Prof.Vineeth N Balasubramanian
Date: Jan 20, 2026 .
Time: 11:30am
Venue: CS-202.

Abstract: he ability to generate patterns while meeting domain constraints is an important problem in the security community, particularly as it enables detecting such patterns while maintaining privacy and trust. Machine learning approaches only succeed in generating such patterns for a short term time horizon or without a guarantee to meet the domain constraints. Logic-guided agents combine concepts from abductive reasoning, logic programming, and rule learning to generate efficient agentic behavior till a long-term horizon as well as adhere to domain constraints. Additionally, supervised models require data that matches the target environment and struggle to adapt to novel environments. Logic-guided agents extend their models with a rule-based error detection and correction framework that learns rules to detect a potential error and correct itself. This lays a foundation for self-correcting agents that abide by domain rules and adapt to unseen domains– a prerequisite for deploying AI systems safely and at scale in the real world. Related papers and videos:

  1. Geospatial Trajectory Generation via Efficient Abduction, video (published at ICLP 2024)
  2. Sea-cret Agents: Maritime Abduction for Region Generation to Expose Dark Vessel Trajectories, video (published at AAMAS 2025)
  3. Rule-Based Error Detection and Correction to Operationalize Movement Trajectory Classification (published at IJCAI 2025)