GRAIN models reasoning as a semantic parsing and tool-execution pipeline, guided by a Structure Invariance Reward, which forces the LLM to learn robust text-to-structure mappings rather than memorizing linguistic artifacts.
Abstract
Despite their potential in standardized graph tasks, Large Language Models (LLMs) remain brittle to real-world shifts in node identifiers and task formulation. While deterministic graph tools are invariant to such shifts, extracting topological structures from noisy text is highly fragile for LLMs, which often overfit to surface patterns. Moreover, mitigating these parsing failures via multi-agent systems incurs prohibitive latency. To address this, we propose GRAIN, a single-agent framework optimized via reinforcement learning. GRAIN models reasoning as a semantic parsing and tool-execution pipeline, guided by a Structure Invariance Reward. By validating extracted intermediate graphs against ground-truth topologies, this reward forces the LLM to learn robust text-to-structure mappings rather than memorizing linguistic artifacts. We also introduce GRIT, a benchmark evaluating sensitivity to such linguistic shifts. GRAIN outperforms multi-agent baselines by 16.45\% in accuracy with approximately 24\% lower latency. Furthermore, it demonstrates superior structural generalization, halving the out-of-distribution (OOD) gap of SFT models (from 15.77\% to 7.80\%) and maintaining robustness on large-scale graphs beyond the training distribution.
PoTRE (Poly-Topological Reasoning Ensembles), a heterogeneous framework that decouples inference into four agents that achieves improved reasoning performance using similar or fewer inference tokens compared to heavily scaled homogeneous baselines is introduced.
While large language models (LLMs) have achieved impressive gains in commonsense reasoning, they often fall into “associative shortcuts,” failing to distinguish correct answers from plausible but constraint-violating hard negatives. This reliance on semantic priors rather than specific situational constraints limits their fine-grained reasoning capabilities. To address this issue, we propose EcoReason (Evolved Commonsense Reasoning), a graph-guided evolutionary and negative-aware reinforcement learning (RL) framework. First, we introduce Graph-Guided Data Evolution, an iterative data generation strategy coupled with the student model’s training progress. In each round, we use knowledge graphs (KGs) to identify deceptive sibling concepts and employ a teacher LLM to create constraint-heavy question answering (QA) data targeting the student’s current blind spots. As training progresses, the synthesized curriculum becomes increasingly challenging. Second, we propose Negative-Aware Policy Optimization (NAPO), an RL algorithm built upon Group Relative Policy Optimization (GRPO). NAPO identifies “stubborn negatives,” defined as incorrect options that are repeatedly selected by sampled policies within the same rollout group, and applies stronger penalties to these recurring distractor-specific errors. Experiments show that EcoReason substantially improves LLM commonsense reasoning in complex constraint-heavy scenarios.
Xin Guan, Jiuxin Cao, Biwei Cao et al.· IEEE Transactions on Audio,...· 0 citations
A novel framework, \textit{GAMER}~(Graph-based Action-centric Memory with Episodic Reasoning), that bridges the gap between inference scaling and episodic memory by decoupling the memory mechanism from LLMs.
Large Language Models (LLMs) are increasingly asked to reason over structured data such as graphs, yet how reliably they can carry out multi-step graph algorithms in language remains unclear. Existing evaluations tend to use simple tasks on small graphs, to score code generation rather than reasoning over the graph itself, or to fix a single input format. We introduce Graph Theory Bench (GT Bench), a benchmark covering 24 classical graph problems in 44 task-structure settings, with over 100,000 examples across four representations: natural language, structured language, adjacency list, and adjacency matrix. Evaluating eight LLMs on GT Bench shows that accuracy is strongly tied to the input representation, that the best representation shifts with graph density, size, and topology as well as with the model, and that this sensitivity persists, attenuated, in the strongest reasoning models. Building on these observations, we propose the Graph Theory Agent (GTA), which pairs a preference-trained representation selector with plan-and-decompose scaffolding around a frozen executor LLM. GTA lifts Phi-4 from 53.5% to 69.1% on the benchmark's easy split and from 33.0% to 41.5% on its hard split, outperforming eight prompting and agent baselines, and transfers without retraining to GraCoRe and NLGraph. Code for benchmark generation and evaluation: https://github.com/xzx34/GTA. The project homepage is available at https://xzx34.github.io/gta/.
Zi-Xiang Xu, Yan-Bo Wang, Chenxi Wang et al.· 2 citations· ⚡1
Although existing multi-agent Retrieval-Augmented Generation (RAG) systems have demonstrated promise on complex multimodal reasoning tasks, they remain fundamentally limited in reasoning depth and memory structure, suffering from inadequate retrieval and state blindness when answering knowledge-intensive questions. To address these limitations, we propose GraMRAG, a graph memory-guided multi-agent RAG framework that integrates a dynamic multimodal memory graph to enable stable, multi-step multimodal reasoning. We introduce a vision-text bridged reasoning paradigm that unifies multi-scale entity cropping with a ReAct-style visual toolchain, enhancing the long-horizon cross-modal reasoning capability. We further construct a multimodal memory graph that formalizes agent reasoning as a dynamic directed acyclic graph (DAG), explicitly modeling action-observation dependencies to mitigate state blindness and suppress redundant retrieval. Moreover, we propose Topology-Aware Policy Optimization (TAPO) that leverages graph topology for critical path identification and targeted node pruning, enabling fine-grained credit assignment across multi-step reasoning trajectories. Extensive experiments on challenging multimodal benchmarks demonstrate that our approach consistently outperforms existing baselines and achieves state-of-the-art performance on complex long-horizon reasoning tasks.
SAGA (Synthetic Agentic Graph Architecture), a system for generating large-scale, semantically rich temporal graphs via a four-phase pipeline, achieves structural realism, semantic richness, and automatic anomaly labeling in a unified framework.
