pith. sign in

arxiv: 2605.26178 · v1 · pith:PIICRFIMnew · submitted 2026-05-25 · 💻 cs.MA · cs.LG

ATOM: Instantiating Budget-Controllable Multi-Agent Collaboration via Nucleus-Electron Hierarchy

Xinkui Zhao , Sai Liu , Yifan Zhang , Qingyu Ma , Zewen Lin , Naibo Wang , Guanjie Cheng , Chang Liu
show 1 more author
Yueshen Xu
This is my paper

Pith reviewed 2026-06-29 19:54 UTC · model grok-4.3

classification 💻 cs.MA cs.LG
keywords multi-agent systemslarge language modelscollaboration topologybudget controlreinforcement learningtoken efficiencynucleus-electron hierarchy
0
0 comments X

The pith

ATOM uses a nucleus-electron hierarchy to make multi-agent LLM collaboration budget-controllable by estimating query difficulty at inference time.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents ATOM as a framework that generates collaboration graphs for LLM-based multi-agent systems while controlling computational budgets. It draws on atomic structure to keep a stable, offline-learned backbone called the nucleus and to activate additional agents called electrons only when needed. A complexity-aware budgeting step estimates how hard each query is from the input and uses that estimate to limit electron creation. The approach is trained with task-driven reinforcement learning so the nucleus learns reliable patterns while electrons adapt per query. Experiments across six benchmarks show the method reaches top performance levels and reduces token consumption by as much as 30 percent relative to prior strong baselines.

Core claim

ATOM instantiates budget-controllable multi-agent collaboration via a nucleus-electron hierarchy: an offline-learned stable collaboration backbone (nucleus) is maintained while query-conditioned agents (electrons) are dynamically activated during inference, with a complexity-aware budgeting strategy that estimates query difficulty from the input alone to strictly regulate electron instantiation.

What carries the argument

Nucleus-electron hierarchy with complexity-aware budgeting strategy that estimates query difficulty to control dynamic agent activation

If this is right

  • Multi-agent systems can separate stable collaboration patterns from query-specific additions without retraining the entire structure each time.
  • Resource use becomes proportional to estimated task demands rather than fixed in advance.
  • Token consumption decreases while benchmark scores remain at or above prior state-of-the-art levels across varied tasks.
  • Reinforcement learning can be applied offline to learn the nucleus while inference-time rules handle electron activation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same nucleus-electron split could be applied to non-LLM agent systems where some coordination rules are fixed and others vary with context.
  • If difficulty estimation proves accurate on new domains, the framework offers a route to automatic scaling of agent teams without manual budget setting.
  • Failures in collaboration might become easier to diagnose by checking whether the nucleus alone suffices or whether the budgeting rule blocked needed electrons.

Load-bearing premise

The budgeting strategy can reliably estimate query difficulty from the input alone and use that estimate to strictly regulate electron instantiation without harming overall performance or stability.

What would settle it

A direct test would measure whether performance drops on held-out queries when the number of electrons is capped according to the model's difficulty estimate but the actual token demand or required agents exceeds that cap.

Figures

Figures reproduced from arXiv: 2605.26178 by Chang Liu, Guanjie Cheng, Naibo Wang, Qingyu Ma, Sai Liu, Xinkui Zhao, Yifan Zhang, Yueshen Xu, Zewen Lin.

Figure 1
Figure 1. Figure 1: Comparison of MAS topology design paradigms. Large language model (LLM)-based agents show strong capabilities across diverse do￾mains [19, 42, 7, 33, 16, 11, 41, 36, 44], yet single agents struggle with complex problems due to limited expertise and reasoning depth [8, 28, 15]. This has driven the shift toward multi-agent systems (MAS), which leverage collective intelli￾gence through specialized roles [13, … view at source ↗
Figure 2
Figure 2. Figure 2: Performance under different agent budgets across difficulty levels. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Pipeline of ATOM topology generation. 3 ATOM for MAS Topology Design To overcome the inherent stability-extensibility trade-off in existing architectures, ATOM employs a two-tier nucleus–electron hierarchy. Specifically, we partition the global agent pool into a persistent, offline-learned nucleus backbone (Vnuc) and a dynamic electron reservoir (Velec). Crucially, this backbone refers strictly to a fixed … view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of performance and token consumption across baselines. Each point is an [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance and token consumption comparison using [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Robustness analysis [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
read the original abstract

Large Language Model (LLM)-based multi-agent systems rely on optimized collaboration topologies to balance performance and communication costs. However, current methods struggle with the inherent stability-extensibility trade-off and often misalign computational budgets with query difficulty. We propose \textsc{ATOM}, an adaptive framework that generates budget-controllable collaboration graphs via a novel task-driven reinforcement learning paradigm. Inspired by atomic structures, \textsc{ATOM} employs a nucleus-electron hierarchy: it maintains a stable, offline-learned collaboration backbone (the nucleus) while dynamically activating query-conditioned agents (electrons) during inference. Crucially, a complexity-aware budgeting strategy aligns resource consumption with task demands by estimating query difficulty to strictly regulate electron instantiation. Extensive experiments across six diverse benchmarks demonstrate that \textsc{ATOM} achieves state-of-the-art performance while improving token efficiency by up to $30\%$ compared to strong baselines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper proposes ATOM, a multi-agent LLM framework using a nucleus-electron hierarchy: a stable offline-learned collaboration backbone (nucleus) paired with dynamically instantiated query-conditioned agents (electrons). It introduces a task-driven RL paradigm to generate budget-controllable graphs and a complexity-aware budgeting strategy that estimates query difficulty from the input to strictly regulate electron count. Experiments across six benchmarks are reported to achieve SOTA performance with up to 30% token-efficiency gains over strong baselines.

Significance. If the budgeting estimator reliably maps input features to true reasoning depth and the RL objective produces stable graphs without hidden parameter dependence, the nucleus-electron separation could offer a practical solution to the stability-extensibility trade-off while delivering measurable efficiency gains. The explicit separation of offline backbone from online instantiation is a clear architectural contribution if the empirical claims are reproducible.

major comments (2)
  1. [Abstract] Abstract: the central SOTA + 30% token-efficiency claim is presented without any description of the baselines, number of runs, statistical tests, or controls for prompt length and model size; this information is required to evaluate whether the efficiency gain is attributable to the budgeting strategy rather than experimental setup.
  2. [Abstract] Abstract: the complexity-aware budgeting strategy is asserted to 'estimate query difficulty to strictly regulate electron instantiation,' yet no equation, feature set, training signal, or ablation is supplied for the estimator; because this mapping is the load-bearing mechanism for both the efficiency gain and the performance-stability alignment, its absence prevents verification that the reported numbers are not the result of post-hoc tuning or over-instantiation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment point by point below and indicate the revisions we will undertake.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central SOTA + 30% token-efficiency claim is presented without any description of the baselines, number of runs, statistical tests, or controls for prompt length and model size; this information is required to evaluate whether the efficiency gain is attributable to the budgeting strategy rather than experimental setup.

    Authors: We agree that the abstract would benefit from additional context to substantiate the performance claims. In the revised manuscript we will expand the abstract to briefly identify the strong baselines, state that results are averaged over multiple runs with statistical testing, and confirm that experiments controlled for prompt length and model size. These controls are already detailed in the experimental section; their mention in the abstract will help readers attribute gains to the budgeting strategy. revision: yes

  2. Referee: [Abstract] Abstract: the complexity-aware budgeting strategy is asserted to 'estimate query difficulty to strictly regulate electron instantiation,' yet no equation, feature set, training signal, or ablation is supplied for the estimator; because this mapping is the load-bearing mechanism for both the efficiency gain and the performance-stability alignment, its absence prevents verification that the reported numbers are not the result of post-hoc tuning or over-instantiation.

    Authors: The referee is correct that the abstract itself supplies none of the technical specifications for the estimator. The full manuscript presents the estimator's equation, input features, RL-derived training signal, and supporting ablations in the methods and experiments sections. To address the concern, we will revise the abstract to include a concise reference to these elements and their grounding in the task-driven RL paradigm. If the main-text description requires further elaboration or additional ablations, we will incorporate them during revision. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation remains self-contained

full rationale

The paper introduces ATOM as a framework using a nucleus-electron hierarchy and a complexity-aware budgeting strategy within a task-driven RL paradigm. The provided abstract and description contain no equations, parameter-fitting steps, or self-citations that reduce any claimed prediction or result to its inputs by construction. The budgeting mechanism is described as estimating difficulty to regulate electrons, but this is presented as an empirical alignment technique rather than a definitional or fitted tautology. Central performance claims rest on external benchmark evaluations, which are independent of any internal derivation chain. No load-bearing step matches the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Based solely on the abstract, the central claim rests on the unverified effectiveness of the nucleus-electron split and the accuracy of the difficulty estimator; no free parameters, axioms, or independent evidence for the invented hierarchy are supplied.

invented entities (2)
  • nucleus (stable offline-learned collaboration backbone) no independent evidence
    purpose: provides a fixed, stable core for collaboration
    Introduced in the abstract as the stable component of the hierarchy.
  • electrons (query-conditioned agents) no independent evidence
    purpose: dynamically activated based on estimated query difficulty
    Introduced in the abstract as the extensible component of the hierarchy.

pith-pipeline@v0.9.1-grok · 5710 in / 1110 out tokens · 30101 ms · 2026-06-29T19:54:51.567967+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 23 canonical work pages · 10 internal anchors

  1. [1]

    Shuowei Cai, Yansong Ning, and Hao Liu. 2025. Agentbalance: Backbone-then-topology design for cost-effective multi-agent systems under budget constraints.arXiv preprint arXiv:2512.11426

    work page arXiv 2025

  2. [2]

    Guangyao Chen, Siwei Dong, Yu Shu, Ge Zhang, Jaward Sesay, Börje F Karlsson, Jie Fu, and Yemin Shi

  3. [3]

    Autoagents: A framework for automatic agent generation.arXiv preprint arXiv:2309.17288

    work page arXiv

  4. [4]

    Hongjiang Chen, Xin Zheng, Yixin Liu, Pengfei Jiao, Shiyuan Li, Huan Liu, Zhidong Zhao, Ziqi Xu, Ibrahim Khalil, and Shirui Pan. 2026. Goagent: Group-of-agents communication topology generation for llm-based multi-agent systems.arXiv preprint arXiv:2603.19677

    work page arXiv 2026

  5. [5]

    Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, and 1 others. 2021. Evaluating large language models trained on code.Preprint, arXiv:2107.03374

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  6. [6]

    Yongchao Chen, Jacob Arkin, Yang Zhang, Nicholas Roy, and Chuchu Fan. 2024. Scalable multi- robot collaboration with large language models: Centralized or decentralized systems? In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 4311–4317. IEEE

    2024

  7. [7]

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. Training verifiers to solve math word problems.Preprint, arXiv:2110.14168

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  8. [8]

    Yihong Dong, Xue Jiang, Zhi Jin, and Ge Li. 2024. Self-collaboration code generation via chatgpt.ACM Transactions on Software Engineering and Methodology, 33(7):1–38

    2024

  9. [9]

    Yilun Du, Shuang Li, Antonio Torralba, Joshua B Tenenbaum, and Igor Mordatch. 2023. Improving factuality and reasoning in language models through multiagent debate. InForty-first International Conference on Machine Learning

    2023

  10. [10]

    Yijia Fan, Jusheng Zhang, Kaitong Cai, Jing Yang, Chengpei Tang, Jian Wang, and Keze Wang. 2026. Cost-effective communication: An auction-based method for language agent interaction. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 29412–29420

    2026

  11. [11]

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021. Measuring massive multitask language understanding.Preprint, arXiv:2009.03300

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  12. [12]

    Sirui Hong, Yizhang Lin, Bang Liu, Bangbang Liu, Binhao Wu, Ceyao Zhang, Danyang Li, Jiaqi Chen, Jiayi Zhang, Jinlin Wang, and 1 others. 2025. Data interpreter: An llm agent for data science. InFindings of the Association for Computational Linguistics: ACL 2025, pages 19796–19821

    2025

  13. [13]

    Sirui Hong, Mingchen Zhuge, Jiaqi Chen, Xiawu Zheng, Yuheng Cheng, Ceyao Zhang, Jinlin Wang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, Chenyu Ran, Lingfeng Xiao, Chenglin Wu, and Jürgen Schmidhuber. 2024. Metagpt: Meta programming for a multi-agent collaborative framework. Preprint, arXiv:2308.00352

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  14. [14]

    Md Ashraful Islam, Mohammed Eunus Ali, and Md Rizwan Parvez. 2024. Mapcoder: Multi-agent code generation for competitive problem solving.arXiv preprint arXiv:2405.11403

    work page arXiv 2024

  15. [15]

    Zhao Kaiya, Michelangelo Naim, Jovana Kondic, Manuel Cortes, Jiaxin Ge, Shuying Luo, Guangyu Robert Yang, and Andrew Ahn. 2023. Lyfe agents: Generative agents for low-cost real-time social interactions. arXiv preprint arXiv:2310.02172. 10

    work page arXiv 2023

  16. [16]

    Zixuan Ke, Fangkai Jiao, Yifei Ming, Xuan-Phi Nguyen, Austin Xu, Do Xuan Long, Minzhi Li, Chengwei Qin, Peifeng Wang, Silvio Savarese, and 1 others. 2025. A survey of frontiers in llm reasoning: Inference scaling, learning to reason, and agentic systems.arXiv preprint arXiv:2504.09037

    work page arXiv 2025

  17. [17]

    Boyan Li, Yuyu Luo, Chengliang Chai, Guoliang Li, and Nan Tang. 2024. The dawn of natural language to sql: Are we fully ready?Proceedings of the VLDB Endowment, 17(11):3318–3331

    2024

  18. [18]

    Shiyuan Li, Yixin Liu, Qingsong Wen, Chengqi Zhang, and Shirui Pan. 2025. Assemble your crew: Automatic multi-agent communication topology design via autoregressive graph generation.arXiv preprint arXiv:2507.18224

    work page arXiv 2025

  19. [19]

    Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. 2017. Program induction by rationale generation: Learning to solve and explain algebraic word problems.Preprint, arXiv:1705.04146

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  20. [20]

    Junwei Liu, Kaixin Wang, Yixuan Chen, Xin Peng, Zhenpeng Chen, Lingming Zhang, and Yiling Lou. 2024. Large language model-based agents for software engineering: A survey.arXiv preprint arXiv:2409.02977

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  21. [21]

    Yixin Liu, Guibin Zhang, Kun Wang, Shiyuan Li, and Shirui Pan. 2025. Graph-augmented large language model agents: Current progress and future prospects.IEEE Intelligent Systems

    2025

  22. [22]

    Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. 2023. Generative agents: Interactive simulacra of human behavior. InProceedings of the 36th annual acm symposium on user interface software and technology, pages 1–22

    2023

  23. [23]

    Arkil Patel, Satwik Bhattamishra, and Navin Goyal. 2021. Are nlp models really able to solve simple math word problems?Preprint, arXiv:2103.07191

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  24. [24]

    Chen Qian, Wei Liu, Hongzhang Liu, Nuo Chen, Yufan Dang, Jiahao Li, Cheng Yang, Weize Chen, Yusheng Su, Xin Cong, and 1 others. 2023. Chatdev: Communicative agents for software development. arXiv preprint arXiv:2307.07924

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  25. [25]

    Chen Qian, Zihao Xie, YiFei Wang, Wei Liu, Kunlun Zhu, Hanchen Xia, Yufan Dang, Zhuoyun Du, Weize Chen, Cheng Yang, Zhiyuan Liu, and Maosong Sun. 2025. Scaling large language model-based multi-agent collaboration.Preprint, arXiv:2406.07155

    work page arXiv 2025

  26. [26]

    Subhro Roy and Dan Roth. 2016. Solving general arithmetic word problems.Preprint, arXiv:1608.01413

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  27. [27]

    Xu Shen, Yixin Liu, Yiwei Dai, Yili Wang, Rui Miao, Yue Tan, Shirui Pan, and Xin Wang. 2025. Understanding the information propagation effects of communication topologies in llm-based multi-agent systems. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing

    2025

  28. [28]

    Chunhao Tian, Yutong Wang, Xuebo Liu, Zhexuan Wang, Liang Ding, Miao Zhang, and Min Zhang. 2025. Agentinit: Initializing llm-based multi-agent systems via diversity and expertise orchestration for effective and efficient collaboration. InFindings of the Association for Computational Linguistics: EMNLP 2025, pages 11870–11902

    2025

  29. [29]

    Karthik Valmeekam, Matthew Marquez, Sarath Sreedharan, and Subbarao Kambhampati. 2023. On the planning abilities of large language models-a critical investigation.Advances in Neural Information Processing Systems, 36:75993–76005

    2023

  30. [30]

    1999.Building the flexible firm: How to remain competitive

    Henk W V olberda. 1999.Building the flexible firm: How to remain competitive. Oxford university press

    1999

  31. [31]

    Wenhui Wang, Furu Wei, Li Dong, Hangbo Bao, Nan Yang, and Ming Zhou. 2020. Minilm: Deep self-attention distillation for task-agnostic compression of pre-trained transformers.Advances in neural information processing systems, 33:5776–5788

    2020

  32. [32]

    Zhexuan Wang, Yutong Wang, Xuebo Liu, Liang Ding, Miao Zhang, Jie Liu, and Min Zhang. 2025. Agentdropout: Dynamic agent elimination for token-efficient and high-performance llm-based multi-agent collaboration.Preprint, arXiv:2503.18891

    work page arXiv 2025

  33. [33]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models.Advances in neural information processing systems, 35:24824–24837

    2022

  34. [34]

    Liwenhan Xie, Chengbo Zheng, Haijun Xia, Huamin Qu, and Chen Zhu-Tian. 2024. Waitgpt: Monitoring and steering conversational llm agent in data analysis with on-the-fly code visualization. InProceedings of the 37th Annual ACM Symposium on User Interface Software and Technology, pages 1–14. 11

    2024

  35. [35]

    Liming Yang, Junyu Luo, Xuanzhe Liu, Yiling Lou, and Zhenpeng Chen. 2025. Bamas: Structuring budget-aware multi-agent systems.arXiv preprint arXiv:2511.21572

    work page arXiv 2025

  36. [36]

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Tom Griffiths, Yuan Cao, and Karthik Narasimhan

  37. [37]

    Tree of thoughts: Deliberate problem solving with large language models.Advances in neural information processing systems, 36:11809–11822

  38. [38]

    Murong Yue. 2025. A survey of large language model agents for question answering.arXiv preprint arXiv:2503.19213

    work page arXiv 2025

  39. [39]

    Guibin Zhang, Yanwei Yue, Zhixun Li, Sukwon Yun, Guancheng Wan, Kun Wang, Dawei Cheng, Jef- frey Xu Yu, and Tianlong Chen. 2025. Cut the crap: An economical communication pipeline for llm-based multi-agent systems. InInternational Conference on Learning Representations

    2025

  40. [40]

    Guibin Zhang, Yanwei Yue, Xiangguo Sun, Guancheng Wan, Miao Yu, Junfeng Fang, Kun Wang, Tianlong Chen, and Dawei Cheng. 2025. G-designer: Architecting multi-agent communication topologies via graph neural networks. InInternational Conference on Machine Learning

    2025

  41. [41]

    Jiayi Zhang, Jinyu Xiang, Zhaoyang Yu, Fengwei Teng, Xionghui Chen, Jiaqi Chen, Mingchen Zhuge, Xin Cheng, Sirui Hong, Jinlin Wang, Bingnan Zheng, Bang Liu, Yuyu Luo, and Chenglin Wu. 2025. Aflow: Automating agentic workflow generation.Preprint, arXiv:2410.10762

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  42. [42]

    Yifan Zhang, Xinkui Zhao, Zuxin Wang, Zhengyi Zhou, Guanjie Cheng, Shuiguang Deng, and Jianwei Yin. 2025. Sortinghat: Redefining operating systems education with a tailored digital teaching assistant. In Companion Proceedings of the ACM on Web Conference 2025, pages 2951–2954

    2025

  43. [43]

    Xinkui Zhao, Zuxin Wang, Yifan Zhang, Guanjie Cheng, Yueshen Xu, Shuiguang Deng, Chang Liu, Naibo Wang, and Jianwei Yin. 2025. Video-qtr: Query-driven temporal reasoning framework for lightweight video understanding.arXiv preprint arXiv:2512.09354

    work page arXiv 2025

  44. [44]

    Li Zhong, Zilong Wang, and Jingbo Shang. 2024. Debug like a human: A large language model debugger via verifying runtime execution step by step. InFindings of the Association for Computational Linguistics ACL 2024, pages 851–870

    2024

  45. [45]

    Han Zhou, Xingchen Wan, Ruoxi Sun, Hamid Palangi, Shariq Iqbal, Ivan Vuli ´c, Anna Korhonen, and Sercan Ö. Arıkk. 2025. Multi-agent design: Optimizing agents with better prompts and topologies. Preprint, arXiv:2502.02533

    work page arXiv 2025

  46. [46]

    Jun-Peng Zhu, Peng Cai, Kai Xu, Li Li, Yishen Sun, Shuai Zhou, Haihuang Su, Liu Tang, and Qi Liu. 2024. Autotqa: Towards autonomous tabular question answering through multi-agent large language models. Proceedings of the VLDB Endowment, 17(12):3920–3933

    2024

  47. [47]

    Mingchen Zhuge, Wenyi Wang, Louis Kirsch, Francesco Faccio, Dmitrii Khizbullin, and Jürgen Schmid- huber. 2024. Gptswarm: Language agents as optimizable graphs. InForty-first International Conference on Machine Learning. 12

    2024