NeSyFS: A Neuro-symbolic Fast-Slow Thinking Framework for LLM Agent under Partial Observability
Duo Xu, Faramarz Fekri
Why It Matters
What makes this one worth your time
This framework addresses critical challenges in deploying LLMs as autonomous agents, particularly in environments with limited information, making it relevant for applications in self-reflection and scientific discovery.
NeSyFS enhances LLM agents' decision-making under uncertainty through a neuro-symbolic approach.
Summary
The paper introduces a neuro-symbolic fast-slow thinking framework (NeSyFS) for large language model agents to effectively operate under partial observability, utilizing a knowledge graph for belief state representation and incorporating both reactive and planning modules.
Key contributions
- Development of the NeSyFS framework for LLM agents.
- Implementation of a knowledge graph to represent belief states.
- Introduction of a reflection module to manage task objective alignment.
Notable insights
- The integration of a knowledge graph for belief state representation is a clever way to enhance context awareness in decision-making.
- The use of a reflection module to switch between fast and slow thinking based on action success is an innovative approach to improve task alignment.
Possible limitations
- Not stated in the abstract.
Abstract
arXiv:2607.28942v2 Announce Type: replace Abstract: Recently Large Language Models (LLMs) have been increasingly deployed as autonomous agents in applications such as self-reflection, retrieval-augmented generation, and scientific discovery. In these settings, agents must act based on limited observations rather than full environmental states, leading to partial observability. This introduces several key challenges: belief state inference, task objective misalignment, and planning under uncertainty. Prior approaches typically condition actions on full or summarized action-observation histories whose redundant and irrelevant information can mislead the decision making of LLM agent. Inspired by human cognition, we propose a novel neuro-symbolic fast-slow thinking (NeSyFS) framework for LLM agent, addressing the challenges introduced by partial observability in a unified approach. We use a knowledge graph (KG) to represent the belief state, providing triplets as context for every module of NeSyFS. The fast-thinking module performs reactive action, while slow-thinking conducts a new uncertainty-aware planning by following the high-level structure of twisted sequential Monte Carlo (TSMC) algorithm. To mitigate the misalignment of task objective, a reflection module is used to reflect fast-thinking actions, and also switches to the slow-thinking module whenever reactive actions repeatedly fail. Experiments on three representative benchmarks, i.e. ALFWorld, Webshop, and ScienceWorld, demonstrate significant advantages over previous methods.