cs.AIOct 4, 2026

AECG: Asymmetric Experience Consolidation and Governance In Multi-Agent Systems

Authors: Ao Tian, Jialong Liu, Daqi Zheng, Xin Sun, Mengting Li, Zhizhao Xiao, Zijian Huang, Honglei Wang, +2 more

Organizations: Beihang University · Wuhan University · ModelBest · Tsinghua University

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Large language model (LLM)-based multi-agent systems increasingly rely on memory to transform execution trajectories into reusable procedural knowledge. Yet repeated retrieval also makes memory errors persistent: memory pollution arises when outdated, weakly supported, or spuriously successful procedures become recurring components of future reasoning. Multi-agent execution introduces an additional structural risk. Scope collapse occurs when procedural knowledge escapes the coordination scope in which it was shown effective and is repeatedly reused at incompatible decision levels, allowing local errors to influence cascades of downstream decisions. Meanwhile, task-level failures provide ambiguous supervision because they rarely reveal which recalled knowledge was responsible. We introduce AECG, a framework for asymmetric experience consolidation and governance for multi-agent systems. AECG turns memory from static experience storage into a dynamic reliability-governance loop, preserving coordination scope and using multi-scale, confidence-aware reliability to detect degradation. It then combines degradation with downstream impact to prioritize high-risk knowledge under a bounded review budget, applies targeted interventions, and reactivates revised skills only after paired replay. Across three multi-agent frameworks and four benchmarks, AECG achieves the best score in 11 of 12 framework--benchmark settings and improves over the strongest competing memory method by as much as 10.23 percentage points; removing scope preservation reduces accuracy by up to 16.89 points. AECG thereby reframes multi-agent memory from passive accumulation into auditable reliability governance. Code is available at https://github.com/fenhg297/AECG

Figures & tables

Figure 1: Conceptual comparison of memory maintenance: coarse accumulation or reactive revision risks persistent errors and utility loss, whereas AECG targets governance for reliable reuse.
Figure 2: Overview of AECG. Agents retrieve scope-aligned procedural memories and skills. Positive evidence consolidates online, while ambiguous failures await qualified local feedback as BadCases. Scope-preserving promotion yields versioned skills, whose dual-timescale reliability and downstream exposure drive budgeted review and replay-gated governance.
MASMemoryPDDLFEVERSciWorldMBPP-PlusAvg.
AutoGenNo-Memory36.5271.4258.6274.5963.43
MacNet-M31.7863.9157.7876.2360.97
MemoryBank40.1268.1756.7075.4163.01
ConMem40.2369.8455.5673.7762.66
G-Memory38.3367.6160.0277.0563.87
AECG46.67 (+6.44)74.49 (+3.07)63.35 (+3.33)82.78 (+5.73)69.79 (+5.92)
Table 1: Performance comparison across frameworks and memory methods, averaged over 3 seeds. Bold indicates the best result, with parentheses showing the absolute percentage-point margin over the runner-up. AECG rows are shaded light green.
VariantSciWorldMBPP+
Full AECG68.1877.87
w/o Scope51.29 (-16.89)69.53 (-8.34)
w/o Dual-Time63.42 (-4.76)71.61 (-6.26)
w/o Exposure64.82 (-3.36)75.29 (-2.58)
Table 2: AgentNet ablations. Parentheses show drops from AECG.
MethodSuccess ↑\uparrowCross-scope ↓\downarrowExposure ↓\downarrow
Hard-scope68.180.0–
Flat-scope51.2936.514.8
Table 3: Scope preservation on AgentNet–ScienceWorld.
Figure 3: Detection and recovery after controlled skill corruption.
Figure 4: Triage under the same confidence gate and fixed review budget.
Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Figure 5: Performance landscape across frameworks, benchmarks, and memory methods. The outlined AECG row corresponds to Table 1 .

Explore similar work

CardsList
Sep 14, 2026cs.AI

CoMem: Collective-Individual Memory Synergy for Evolutionary Multi-Agent Systems

Designing effective memory mechanisms is crucial for advancing LLM-driven Multi-Agent Systems (MAS), helping agents learn together and perform better over time. While recent work has led to strong cooperation skills, most methods still use flat, unstructured memories, which easily get filled with noise and erase differences between agents. To address this, we introduce the concept of collective-individual memory synergy and propose CoMem, an architecture that unifies both private experience and shared knowledge for multi-agent learning. CoMem features:(i) Private Experience Sedimentation, which lets each agent keep and update its own useful memories over time;(ii) Collective Wisdom Curation, which carefully selects only widely proven ideas to be shared among agents;(iii)Parallel Dual-Stream Retrieval, which allows agents to draw both from their own memory and the group's wisdom, using clustering to ensure diversity.Experiments on ALFWorld and PDDL benchmarks show that CoMem achieves strong overall performance and robustly avoids memory pollution.

Chengxin Yu, Zhaoxin Fan, Faguo Wu +3

Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing, School of Artificial Intelligence, Beihang University · Beijing academy of blockchain and edge computing · National University of Defense Technology +1

Multi-Agent Evolution
AskPDF
Jul 30, 2026cs.MA

ΣΣ-Mem: An Online Reliability Memory for LLM-based Multi-Agent Systems

Memory is central to long-horizon LLM agents, yet existing memory systems primarily preserve interaction content rather than modeling which agents can be trusted and under what conditions. This limitation is particularly important in multi-agent systems, where a central model may be unable to directly verify plausible or correlated peer responses. We introduce ΣΣ-Mem, an online reliability memory that records historical competence evidence for individual peers and peer relationship evidence across the peer set. Both forms of evidence are maintained as real symmetric states and updated from post-decision correctness feedback. By Weyl's inequality, the spectral change caused by each event-level update is bounded, enabling stable online adaptation without retraining the underlying models. ΣΣ-Mem provides a general write-and-read interface: the same memory can be used for residual steering of a central model, response-free peer routing, or reliability-weighted voting. Across five Qwen-family models, ΣΣ-Mem adapts to counterfactual reliability shifts and generalizes to unseen peers and task domains. Direct memory readouts also outperform majority voting and the best fixed peer over the full OOD evaluation set. Moreover, performance improves consistently as more correctness feedback becomes available, indicating that ΣΣ-Mem progressively accumulates actionable reliability information. These results establish reliability memory as a reusable foundation for adaptive coordination in LLM-based multi-agent systems.

Peilin Feng, Suorong Yang, Soujanya Poria

1DeCLaRe Lab, Nanyang Technological University

Multi-Agent Large Language Model SystemsPeers
AskPDF
Jan 9, 2026cs.AI

StackPlanner: A Centralized Hierarchical Multi-Agent System with Task-Experience Memory Management

Multi-agent systems based on large language models, particularly centralized architectures, have recently shown strong potential for complex and knowledge-intensive tasks. However, central agents often suffer from unstable long-horizon collaboration due to the lack of memory management, leading to context bloat, error accumulation, and poor cross-task generalization. To address both task-level memory inefficiency and the inability to reuse coordination experience, we propose StackPlanner, a hierarchical multi-agent framework with explicit memory control. StackPlanner addresses these challenges by decoupling high-level coordination from subtask execution with active task-level memory control, and by learning to retrieve and exploit reusable coordination experience via structured experience memory and reinforcement learning. Experiments on multiple deep-search and agent system benchmarks demonstrate the effectiveness of our approach in enabling reliable long-horizon multi-agent collaboration.

Ruizhe Zhang, Xinke Jiang, Zhibang Yang +12

School of Computer Science, Peking University, Beijing, China · 2National Engineering Research Center for Software Engineering, Peking University, Beijing, China · 3Key Laboratory of High Confidence Software Technologies, Ministry of Education, Beijing, China +5

Global CoordinationPlanner AgentAgentic BenchmarksPeak Memory
AskPDF