TopoAgent: A Self-Evolving Topological Agent for Multimodal Scientific Reasoning

2026-07-16

AI Summary

This work addresses critical limitations of multimodal large language models in scientific reasoning—namely, vision-semantic misalignment induced by linear planning, hallucinations in long-context scenarios, and rigid task granularity—by proposing a self-evolving topological agent framework. The framework replaces linear reasoning pathways with a dynamic, state-isolated directed acyclic graph (DAG), wherein a front-end decomposer breaks complex problems into visually anchored atomic units and orchestrates their execution according to dependency relations. A runtime dynamic atomic fission mechanism automatically refines bottleneck nodes, while strict context isolation enables an interference-resistant, self-correcting reasoning paradigm. Evaluated across multiple scientific reasoning benchmarks in mathematics, physics, and chemistry, the approach significantly outperforms existing linear-agent architectures, demonstrating superior robustness and accuracy.

Abstract

While Multimodal Large Language Models (MLLMs) excel in general tasks, rigorous scientific reasoning remains challenging due to the limitations of monolithic, linear planning. Such sequential designs often suffer from visual-semantic misalignment, long-context hallucinations, and brittle execution under fixed task granularity. We propose TopoAgent, a self-evolving topological framework that replaces linear trajectories with dynamic, state-isolated graph evolution. TopoAgent first employs a front-end decomposer to fracture complex queries into visually-grounded atoms. These atoms are organized into a Directed Acyclic Graph (DAG) based on their dependencies, enabling strict context isolation to shield the reasoning engine from irrelevant historical noise. Furthermore, we introduce adaptive atomic fission, which dynamically splits bottleneck nodes into finer-grained sub-atoms at runtime when tool capability boundaries are exceeded. Extensive experiments across mathematics, physics, and chemistry benchmarks demonstrate that TopoAgent significantly outperforms state-of-the-art linear agent frameworks, providing a robust, noise-resistant, and self-correcting paradigm for autonomous scientific reasoning.

Problem

• scientific reasoning
• multimodal large language models
• visual-semantic misalignment
• long-context hallucinations
• task granularity

Innovation

• topological reasoning
• self-evolving agent
• directed acyclic graph
• adaptive atomic fission
• multimodal scientific reasoning