Pith. sign in

REVIEW 4 major objections 7 minor 46 references

Tacit Learning with Adaptive Information Selection for Cooperative Multi-Agent Reinforcement Learning

T0 review · 4 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that cooperative MARL agents can learn implicit coordination without communication by regenerating a teammate-information summary from local histories, and that this improves CTDE baselines and beats explicit…

desk verdict A plausible explicit-to-tacit value-decomposition extension whose core regeneration assumption is under-supported, and whose own Table 2 contradicts the plug-and-play claim. read the letter →

arxiv 2412.15639 v2 pith:55ZBTXE2 submitted 2024-12-20 cs.MA cs.AIcs.LG

Lunjun Liu , Weilai Jiang , Yaonan Wang This is my paper · ORCID
classification cs.MAcs.AIcs.LG
keywords multi-agentreinforcementlearningtacitadaptiveinformationselectioncentralizedtrainingdecentralizedexecutionimplicitcommunicationvaluedecompositionselectivestatespacemodelStarCraftChallenge
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes SICA, an add-on architecture for cooperative multi-agent reinforcement learning under the centralized-training/decentralized-execution paradigm. SICA's goal is to let each agent select the most decision-relevant parts of its observation history and to learn a tacit understanding of teammates: during training, an attention mechanism shares hidden states among agents, while a regeneration block learns to reproduce that shared information from a single agent's local history alone. The shared information is blended with the local reconstruction through a cosine-annealed schedule, so at execution time no communication is needed. On SMAC, SMACv2, and Google Research Football, the authors report that SICA improves value-decomposition baselines such as QMIX and VDN and matches or surpasses explicit communication methods.

What carries the argument

The load-bearing mechanism is the train-time/execution-time division of labor around the true information $v_i = \sum_{i\neq j} w_{i,j} h_j$, an attention-weighted combination of other agents' hidden states. During training, the Communication Block provides $v_i$; the Regeneration Block must learn to produce $\hat{v}_i$ from local observation-action history, and the cross-information $\bar{v}_i = (1-\alpha(t))\hat{v}_i + \alpha(t) v_i$ feeds the Q-value. The Selection Block uses a gated MLP plus an S6 selective state-space layer with input-dependent $\Delta$, $B$, and $C$ parameters to filter irrelevant information from a mini-buffer of past observation-action pairs. The cosine annealing of $\alpha(t)$ and the thresholded weighting $\sigma(t)$ of the alignment loss are what make the transition from centralized to decentralized execution smooth rather than abrupt.

What would settle it

Train SICA on a task where each agent's observations are deliberately stripped of any information about teammates, such as agents in separate rooms sharing only a team reward; if the win rate collapses when the training-time communication is annealed away but stays high when the true information is available, then the regeneration assumption fails.

Watch

Extended reading notes

Core claim

The central claim is that the two bottlenecks of CTDE, agents' inability to filter input information and their inability to access global information during execution, can be overcome jointly. SICA adds three blocks to each agent network: a Selection Block (two MLPs and an S6 selective state-space layer) that gates the observation-action history and decides what to remember; a Communication Block that computes an attention-weighted sum of other agents' hidden states to form the true information $v_i$; and a Regeneration Block that maps the agent's own history to an estimate $\hat{v}_i$. The alignment loss $\mathcal{L}_{Align}$ drives $\hat{v}_i$ toward $v_i$, while the cross-information blend uses a coefficient $\alpha(t)$ annealed from 1 to 0, so the architecture gradually stops relying on the communicated target. The paper reports that this yields better win rates than QMIX, VDN, QTRAN, QMIX-Attention, and NDQ on hard and super-hard SMAC maps, higher scores than CDS-QMIX and QMIX-Attention on GRF, and consistent superiority in SMACv2; ablations show both the Selection Block and progressive regeneration contribute.

Load-bearing premise

The assumption that carries the method is that a single agent's own local observation-action history contains enough information about its teammates' behavior to reconstruct the shared message, and that this stays true while the agents' policies change during training.

Editorial extensions

If this is right

  • SICA can be dropped into value-decomposition learners such as QMIX and VDN, or actor-critic learners such as MADDPG, and improve them without changing the base algorithm's training objective beyond the added alignment loss.
  • Execution becomes communication-free, so the method avoids message latency and applies where explicit channels are unavailable or too costly.
  • On SMACv2 with larger team sizes, SICA outperforms the explicit-communication baseline QMIX-Attention, supporting the claim that the selection mechanism helps as the volume of information grows.
  • The gradual annealing from true to regenerated information outperforms both no regeneration (SICA-ZERO) and late switching (SICA-ONE), so the smooth transition is load-bearing for the method.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable implication not explored in the paper is that the alignment loss acts as an information bottleneck: the local history must retain exactly the features of teammates' behavior that the attention target encodes, so measuring the mutual information between $\hat{v}_i$ and $v_i$ along training could reveal when the reconstruction becomes the limiting factor.
  • Because the regeneration target $v_i$ is itself learned through attention weights that shift during training, the method's success may depend on target stability; comparing cosine annealing against a constant small $\alpha$ in tasks with non-stationary teammates would isolate that dependence.
  • The paper evaluates discrete-action benchmarks; a natural next test is continuous-action domains, where the regenerated target would need to represent teammate policies rather than action-value information.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 7 minor

Summary. The paper proposes SICA, a modification of the agent networks used in CTDE value-decomposition MARL methods such as QMIX and VDN. SICA adds three components: a Selection Block (an S6 selective state-space layer with a gating unit and a mini-buffer of recent observation-action pairs), a Communication Block (attention over agents' hidden states to produce a 'true information' vector during training), and a Regeneration Block (an MLP that reconstructs the true information from a single agent's local history). A cosine-annealed weight alpha interpolates between the true information and the regenerated estimate during training, reaching zero at inference so that execution is communication-free. The loss is the sum of the standard TD loss and an alignment loss whose weight follows a threshold schedule. Experiments on SMAC, SMACv2, and GRF are reported against VDN, QMIX, QTRAN, NDQ, QMIX-Attention, CDS-QMIX, and QMIX-CADP, along with ablations for the selection block and for the progressive alpha schedule. The paper also shows an application to VDN and to MADDPG. Section 6 lists limitations, including the fragility of the regeneration block when agents' observations or trajectories are dissimilar and the need to tune several hyperparameters.

Significance. If the core result holds, SICA provides a generic upgrade for CTDE agent networks that removes execution-time communication while retaining train-time coordination, and it adds an information-selection mechanism that is currently absent from standard CTDE methods. The paper deserves credit for evaluating against a broad set of baselines, including explicit communication methods, and for running ablations that isolate the contribution of the selection block and of progressive information regeneration. Those ablations are a genuine asset. However, the significance is materially weakened by the contradiction in Table 2, the qualitative-only evidence for the central regeneration mechanism, and the absence of code and hyperparameter details. The three-seed confidence intervals are too thin to support the strongly worded 'consistently outperforms' claims. The central value-decomposition comparisons for QMIX and VDN are internally coherent, but the broader plug-and-play generality claim is not established.

major comments (4)
  1. [Section 5.3, Table 2] The claim that 'SICA continues to enhance the overall performance of the framework' is directly contradicted by the table: SICA-MADDPG has lower mean return than MADDPG in all three environments (Predator-Prey 725.11 vs 831.48, Navigation -255.23 vs -233.31, Pantomime -926.55 vs -812.79) and substantially larger standard deviations. This undermines the plug-and-play generality claim in the Abstract and Section 1. The authors should either remove the MADDPG claim, restrict the plug-and-play claim to value-decomposition methods, or explain why the actor-critic setting is outside the intended scope.
  2. [Section 4.3, Eq. 12; Section 6] The regeneration mechanism is the load-bearing step for decentralized execution, but it is validated only qualitatively. The target v_i in Eq. 12 is the attention-weighted sum of all other agents' hidden states (Eqs. 7-8), while the regressor \hat v_i uses only agent i's local observation-action history. In stochastic environments such as SMACv2, multiple joint hidden states can be consistent with the same local history, making the regression target ill-posed, and the alignment loss minimizes only an expected squared error, not pointwise accuracy. Since alpha(t)=0 at inference (Eq. 10), the entire decentralized Q-value depends on this reconstruction. Figure 10 shows a t-SNE visualization but reports no numerical regeneration error, no per-agent breakdown, and no sensitivity to the alpha schedule or alignment weights. The paper needs quantitative evidence that the reconstruction remains accurate under the policy shift induced by the alpha schedule, or it should temper the claim that SICA 'relies solely on local information' without loss of coordination quality.
  3. [Reproducibility] The manuscript does not provide code or hyperparameter values. The free parameters include the mini-buffer capacity b, the alignment loss weights beta_1 and beta_2 and threshold T in Eq. 14, the alpha schedule endpoints in Eq. 10, network dimensions, learning rates, and the unspecified schedule by which the Regeneration Block's mini-buffer is reduced from other agents' observation-action pairs to zero (Section 4.2). Without these details, the experiments cannot be reproduced or independently checked, and the claim of 'seamless integration' with any value decomposition method is not verifiable. At minimum, a supplementary table of hyperparameters and a statement about which SMAC/SMACv2 version and reward settings were used are necessary.
  4. [Section 5, statistical support] All learning curves are reported as means over three seeds with 95% confidence intervals. With three seeds, the confidence interval is highly sensitive to a single outlier and does not support the language 'SICA consistently outperforms the baselines across all maps.' The paper should report the number of runs per cell, use a more robust uncertainty estimate (e.g., interquartile range or individual runs overlaid), or apply a paired statistical test across maps. This issue is directly relevant to the headline comparison claims in Figures 3 and 4.
minor comments (7)
  1. [Section 5.1] There is a typo 'clasSICAl' in the enumeration of baselines; 'classical' is intended.
  2. [Section 4] The word 'decompostition' appears in the sentence 'can be seamlessly integrated with any value decompostition method'; this should be 'decomposition'.
  3. [References] The citation for Google Research Football is given as [9], but reference [9] is 'It's Raw! Audio Generation with State-Space Models'; the GRF paper by Kurach et al. is reference [18]. Also, references [40] and [41] are the same NDQ paper (Learning Nearly Decomposable Value Functions Via Communication Minimization) and should be merged.
  4. [Eq. 6] The notation Split(x_t^i) is not defined, and the intermediate variable z_t^i is introduced without explanation; please clarify what the split is and how the two MLP outputs interact.
  5. [Section 4.2] The mini-buffer is described as containing the preceding b observation-action pairs, but the listed set is {(o_{t-1}^i, u_{t-2}^i), ..., (o_{t-b}^i, u_{t-b-1}^i)}, which pairs each observation with the action from the previous time step; clarify whether this temporal offset is intentional or a typo.
  6. [Figure 2] The caption uses 'the Nth agent' where 'agent i' is meant; the figure is also quite small and the connection between the Regeneration Block and the agent value head is hard to follow.
  7. [Abstract] The abstract states that agents rely 'solely on local information' without adding 'during execution'; since training uses global information through the Communication Block, the phrasing should be tightened to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SICA's regeneration objective is a learned approximation trained with an auxiliary loss, and its headline claims are evaluated on external benchmark win rates.

full rationale

The paper's central mechanism is the Regeneration Block, which learns to approximate the attention-based true information v_i from a single agent's local observation-action history (Eqs. 7-12). This is a standard auxiliary reconstruction objective, not a definitional identity: v_i is computed from other agents' hidden states via the Communication Block, while v_hat_i is produced by a separate Regeneration Block parameterized by different weights. The alignment loss in Eq. 12 trains the approximation but does not make the prediction equal to the target by construction. The headline claim—that SICA outperforms CTDE and explicit-communication baselines—is supported by externally evaluated win rates on SMAC, SMACv2, and GRF (Figures 3-4, Table 1), not by a quantity that was fit to match those benchmarks. The paper cites external prior work (QMIX, Mamba, SMAC, etc.) and invokes no uniqueness theorem or load-bearing self-citation. Section 6 explicitly concedes that the Regeneration Block 'may struggle when agents' observations or trajectories are dissimilar'; this is an acknowledged empirical limitation about identifiability and distribution shift, not a circular step. The auxiliary loss is self-referential in the sense that its target is produced by the same network family being trained, but this is the ordinary bootstrapping nature of RL auxiliary tasks and does not reduce the paper's performance claims to their inputs. Therefore, no circular step can be quoted with a specific equation-to-equation or fit-to-prediction reduction, and the appropriate score is 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on empirical transfer of privileged information, not on a derivation. The load-bearing assumptions are that S6-based selection improves attention, that L_Align can be minimized to a useful level, and that the benchmark evaluation is representative. No code is available to check the latter.

free parameters (3)
  • alpha schedule endpoints = alpha_start=1, alpha_final=0
    Eq. 10 sets the timing of the transition from true information to regenerated information; these values are chosen by hand and no sensitivity analysis is reported.
  • alignment-loss schedule (T, beta_1, beta_2) = not reported
    Eq. 14 weights the auxiliary alignment loss by a threshold T and two beta values; the paper states they are fixed hyperparameters but never gives their values.
  • mini-buffer capacity b = not reported
    Section 4.2 defines the Selection Block input as the past b observation-action pairs; b is a hyperparameter whose value is absent.
assumptions (4)
  • domain assumption The Dec-POMDP formulation and IGM principle from Section 3 are valid models for the evaluated benchmarks.
    The paper inherits these models from prior literature and does not justify them for SMAC/SMACv2/GRF beyond standard practice.
  • ad hoc to paper The true information vector v_i, produced by training-time attention over agents' hidden states, is a learnable function of any single agent's local observation-action history.
    Eq. 12 minimizes the gap between v_hat_i and v_i, and Eq. 9 gradually removes v_i at execution. No proof or sensitivity study shows this transfer is reliable.
  • ad hoc to paper The S6 selective state-space layer acts as an effective information-selection mechanism for MARL observations.
    Section 4.2 and the ablation in Figure 6 assume the Mamba-style S6 layer generalizes to agent observations; only empirical evidence is offered.
  • ad hoc to paper Parameter sharing across agents is acceptable and does not distort the reported gains.
    Section 6 states parameter sharing is used to shorten training; the paper does not report heterogeneous-agent runs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Tacit Learning with Adaptive Information Selection for Cooperative Multi-Agent Reinforcement Learning." pith.science (2026). https://pith.science/paper/55ZBTXE2

@misc{pith2026241215639,
  author       = {Pith},
  title        = {Pith review of: Tacit Learning with Adaptive Information Selection for Cooperative Multi-Agent Reinforcement Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/55ZBTXE2}},
  note         = {Machine review of arXiv:2412.15639}
}
read the original abstract

In multi-agent reinforcement learning (MARL), the centralized training with decentralized execution (CTDE) framework has gained widespread adoption due to its strong performance. However, the further development of CTDE faces two key challenges. First, agents struggle to autonomously assess the relevance of input information for cooperative tasks, impairing their decision-making abilities. Second, in communication-limited scenarios with partial observability, agents are unable to access global information, restricting their ability to collaborate effectively from a global perspective. To address these challenges, we introduce a novel cooperative MARL framework based on information selection and tacit learning. In this framework, agents gradually develop implicit coordination during training, enabling them to infer the cooperative behavior of others in a discrete space without communication, relying solely on local information. Moreover, we integrate gating and selection mechanisms, allowing agents to adaptively filter information based on environmental changes, thereby enhancing their decision-making capabilities. Experiments on popular MARL benchmarks show that our framework can be seamlessly integrated with state-of-the-art algorithms, leading to significant performance improvements.

Figures

Figures reproduced from arXiv: 2412.15639 by the authors.

Figure 1
Figure 1. A case study in selection mechanism. The colored sections signify the information that the agent elects to remember, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The overall framework of SICA. We illustrate the network architecture using the example of the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Performance comparison between SICA and baselines on SMAC. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Performance comparison between SICA and baselines on SMACv2. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Learning curves with different numbers of agents in SMACv2 [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 7
Figure 7. Figure 7: Performance comparison between SICA and SICA’s [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 6
Figure 6. Figure 6: Performance comparison between SICA and ICA. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 9
Figure 9. Figure 9: Performance comparison between SICA and QMIX-CADP. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Visualizations of the regenerated information. [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 23 canonical work pages

  1. [1]

    Christopher Berner, Greg Brockman, Brooke Chan, Vicki Cheung, Przemyslaw Debiak, Christy Dennison, David Farhi, Quirin Fischer, Shariq Hashme, Christo- pher Hesse, Rafal Józefowicz, Scott Gray, Catherine Olsson, Jakub W. Pachocki, Michael Petrov, Henrique Pondé de Oliveira Pinto, Jonathan Raiman, Tim Sali- mans, Jeremy Schlatter, Jonas Schneider, Szymon S...

    arXiv 2019

  2. [2]

    Bernstein, Shlomo Zilberstein, and Neil Immerman

    Daniel S. Bernstein, Shlomo Zilberstein, and Neil Immerman. 2000. The Complex- ity of Decentralized Control of Markov Decision Processes. ArXiv abs/1301.3836 (2000)

    arXiv 2000

  3. [3]

    Caroline Claus and Craig Boutilier. 1998. The Dynamics of Reinforcement Learn- ing in Cooperative Multiagent Systems. In AAAI/IAAI

    work page 1998

  4. [4]

    Rabbat, and Joelle Pineau

    Abhishek Das, Théophile Gervet, Joshua Romoff, Dhruv Batra, Devi Parikh, Michael G. Rabbat, and Joelle Pineau. 2018. TarMAC: Targeted Multi-Agent Communication. In International Conference on Machine Learning

    work page 2018

  5. [5]

    Ziluo Ding, Tiejun Huang, and Zongqing Lu. 2020. Learning Individually Inferred Communication for Multi-Agent Cooperation. ArXiv abs/2006.06455 (2020)

    work page Pith review arXiv 2020

  6. [6]

    Moalla, Mikayel Samvelyan, Mingfei Sun, Anuj Mahajan, Jakob Nicolaus Foerster, and Shimon Whiteson

    Benjamin Ellis, S. Moalla, Mikayel Samvelyan, Mingfei Sun, Anuj Mahajan, Jakob Nicolaus Foerster, and Shimon Whiteson. 2022. SMACv2: An Improved Benchmark for Cooperative Multi-Agent Reinforcement Learning. ArXiv abs/2212.07489 (2022)

    arXiv 2022

  7. [7]

    Foerster, Yannis Assael, Nando de Freitas, and Shimon Whiteson

    Jakob N. Foerster, Yannis Assael, Nando de Freitas, and Shimon Whiteson. 2016. Learning to Communicate with Deep Multi-Agent Reinforcement Learning.ArXiv abs/1605.06676 (2016)

    arXiv 2016

  8. [8]

    Foerster, Gregory Farquhar, Triantafyllos Afouras, Nantas Nardelli, and Shimon Whiteson

    Jakob N. Foerster, Gregory Farquhar, Triantafyllos Afouras, Nantas Nardelli, and Shimon Whiteson. 2017. Counterfactual Multi-Agent Policy Gradients. In AAAI Conference on Artificial Intelligence

    work page 2017

Show all 46 references
  1. [9]

    Karan Goel, Albert Gu, Chris Donahue, and Christopher R’e. 2022. It’s Raw! Audio Generation with State-Space Models. In International Conference on Machine Learning

    2022

  2. [10]

    Albert Gu and Tri Dao. 2023. Mamba: Linear-Time Sequence Modeling with Selective State Spaces. ArXiv abs/2312.00752 (2023)

    2023 arXiv

  3. [11]

    Albert Gu, Tri Dao, Stefano Ermon, Atri Rudra, and Christopher Ré. 2020. HiPPO: Recurrent Memory with Optimal Polynomial Projections. ArXiv abs/2008.07669 (2020)

    2020 arXiv

  4. [12]

    Albert Gu, Karan Goel, and Christopher R’e. 2021. Efficiently Modeling Long Sequences with Structured State Spaces. ArXiv abs/2111.00396 (2021)

    2021 arXiv

  5. [13]

    Ankit Gupta and Jonathan Berant. 2022. Diagonal State Spaces are as Effective as Structured State Spaces. ArXiv abs/2203.14343 (2022)

    2022 arXiv

  6. [14]

    Jian Hu, Siyang Jiang, Seth Austin Harding, Haibin Wu, and Shihua Liao. 2021. Re- thinking the Implementation Tricks and Monotonicity Constraint in Cooperative Multi-Agent Reinforcement Learning

    2021

  7. [15]

    Ortega, DJ Strouse, Joel Z

    Natasha Jaques, Angeliki Lazaridou, Edward Hughes, Çaglar Gülçehre, Pedro A. Ortega, DJ Strouse, Joel Z. Leibo, and Nando de Freitas. 2018. Social Influence as Intrinsic Motivation for Multi-Agent Deep Reinforcement Learning. In Interna- tional Conference on Machine Learning

    2018

  8. [16]

    Jiechuan Jiang and Zongqing Lu. 2018. Learning Attentional Communication for Multi-Agent Cooperation. In Neural Information Processing Systems

    2018

  9. [17]

    Knepper, Christoforos Mavrogiannis, Julia Proft, and Claire Liang

    Ross A. Knepper, Christoforos Mavrogiannis, Julia Proft, and Claire Liang. 2017. Implicit Communication in a Joint Action. 2017 12th ACM/IEEE International Conference on Human-Robot Interaction (HRI (2017), 283–292

    2017

  10. [18]

    Karol Kurach, Anton Raichuk, Piotr Stańczyk, Michal Zajac, Olivier Bachem, Lasse Espeholt, Carlos Riquelme, Damien Vincent, Marcin Michalski, Olivier Bousquet, and Sylvain Gelly. 2019. Google Research Football: A Novel Reinforcement Learning Environment. ArXiv abs/1907.11180 (2019)

    2019 arXiv

  11. [19]

    Chenghao Li, Chengjie Wu, Tonghan Wang, Jun Yang, Qianchuan Zhao, and Chongjie Zhang. 2021. Celebrating Diversity in Shared Multi-Agent Reinforce- ment Learning. In Neural Information Processing Systems

    2021

  12. [20]

    Dapeng Li, Zhiwei Xu, Bin Zhang, and Guoliang Fan. 2023. From Explicit Com- munication to Tacit Cooperation: A Novel Paradigm for Cooperative MARL. ArXiv abs/2304.14656 (2023)

    2023 arXiv

  13. [21]

    Hanxiao Liu, Zihang Dai, David So, and Quoc V Le. 2021. Pay attention to mlps. Advances in neural information processing systems 34 (2021), 9204–9215

    2021

  14. [22]

    Abbeel, and Igor Mordatch

    Ryan Lowe, Yi Wu, Aviv Tamar, Jean Harb, P. Abbeel, and Igor Mordatch. 2017. Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments. ArXiv abs/1706.02275 (2017)

    2017 arXiv

  15. [23]

    David Henry Mguni, Joel Jennings, and Enrique Munoz de Cote. 2018. De- centralised Learning in Systems with Many, Many Strategic Agents. ArXiv abs/1803.05028 (2018)

    2018 arXiv

  16. [24]

    David Henry Mguni, Joel Jennings, Sergio Valcarcel Macua, Emilio Sison, Sofia Ceppi, and Enrique Munoz de Cote. 2019. Coordinating the Crowd: Inducing Desirable Equilibria in Non-Cooperative Systems. In Adaptive Agents and Multi- Agent Systems

    2019

  17. [25]

    Peng Peng, Ying Wen, Yaodong Yang, Quan Yuan, Zhenkun Tang, Haitao Long, and Jun Wang. 2017. Multiagent Bidirectionally-Coordinated Nets: Emergence of Human-level Coordination in Learning to Play StarCraft Combat Games. arXiv: Artificial Intelligence (2017)

    2017

  18. [26]

    Dawei Qiu, Jianhong Wang, Zihang Dong, Yi Wang, and Goran Strbac. 2023. Mean-Field Multi-Agent Reinforcement Learning for Peer-to-Peer Multi-Energy Trading. IEEE Transactions on Power Systems 38 (2023), 4853–4866

    2023

  19. [27]

    Dawei Qiu, Jianhong Wang, Junkai Wang, and Goran Strbac. 2021. Multi-Agent Reinforcement Learning for Automated Peer-to-Peer Energy Trading in Double- Side Auction Market. In International Joint Conference on Artificial Intelligence

    2021

  20. [28]

    Tabish Rashid, Mikayel Samvelyan, C. S. D. Witt, Gregory Farquhar, Jakob N. Foerster, and Shimon Whiteson. 2018. QMIX: Monotonic Value Function Fac- torisation for Deep Multi-Agent Reinforcement Learning. ArXiv abs/1803.11485 (2018)

    2018 arXiv

  21. [29]

    Mikayel Samvelyan, Tabish Rashid, C. S. D. Witt, Gregory Farquhar, Nantas Nardelli, Tim G. J. Rudner, Chia-Man Hung, Philip H. S. Torr, Jakob N. Foerster, and Shimon Whiteson. 2019. The StarCraft Multi-Agent Challenge. ArXiv abs/1902.04043 (2019)

    2019 arXiv

  22. [30]

    Kyunghwan Son, Daewoo Kim, Wan Ju Kang, David Earl Hostallero, and Yung Yi. 2019. QTRAN: Learning to Factorize with Transformation for Cooperative Multi-Agent Reinforcement Learning. ArXiv abs/1905.05408 (2019)

    2019 arXiv

  23. [31]

    Sainbayar Sukhbaatar, Arthur Szlam, and Rob Fergus. 2016. Learning Multiagent Communication with Backpropagation. InNeural Information Processing Systems

    2016

  24. [32]

    Czarnecki, Viní- cius Flores Zambaldi, Max Jaderberg, Marc Lanctot, Nicolas Sonnerat, Joel Z

    Peter Sunehag, Guy Lever, Audrunas Gruslys, Wojciech M. Czarnecki, Viní- cius Flores Zambaldi, Max Jaderberg, Marc Lanctot, Nicolas Sonnerat, Joel Z. Leibo, Karl Tuyls, and Thore Graepel. 2017. Value-Decomposition Networks For Cooperative Multi-Agent Learning. ArXiv abs/1706.0...

    2017 arXiv

  25. [33]

    Ming Tan. 1993. Multi-agent reinforcement learning: Independent vs. cooperative agents. In Proceedings of the tenth international conference on machine learning . 330–337

    1993

  26. [34]

    Zheng Tian, Shihao Zou, Ian Davies, Tim Warr, Lisheng Wu, Haitham Bou- Ammar, and Jun Wang. 2018. Learning to Communicate Implicitly by Actions. In AAAI Conference on Artificial Intelligence

    2018

  27. [35]

    Laurens Van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-SNE. Journal of machine learning research 9, 11 (2008)

    2008

  28. [36]

    Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N

    Ashish Vaswani, Noam M. Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. In Neural Information Processing Systems

    2017

  29. [37]

    Jianhao Wang, Zhizhou Ren, Terry Liu, Yang Yu, and Chongjie Zhang. 2020. QPLEX: Duplex Dueling Multi-Agent Q-Learning. ArXiv abs/2008.01062 (2020)

    2020 arXiv

  30. [38]

    Jianhong Wang, Wangkun Xu, Yunjie Gu, Wenbin Song, and Tim C. Green

  31. [39]

    Jiangxing Wang, Deheng Ye, and Zongqing Lu. 2022. More Centralized Train- ing, Still Decentralized Execution: Multi-Agent Conditional Policy Factorization. ArXiv abs/2209.12681 (2022)

    2022 arXiv

  32. [41]

    Tonghan Wang, Jianhao Wang, Chongyi Zheng, and Chongjie Zhang. 2019. Learning nearly decomposable value functions via communication minimization. arXiv preprint arXiv:1910.05366 (2019)

    2019 arXiv

  33. [42]

    Zhiwei Xu, Bin Zhang, Dapeng Li, Zeren Zhang, Guangchong Zhou, and Guoliang Fan. 2022. Consensus Learning for Cooperative Multi-Agent Reinforcement Learning. In AAAI Conference on Artificial Intelligence

    2022

  34. [43]

    Tianhao Zhang, Yueheng Li, Chen Wang, Guangming Xie, and Zongqing Lu

  35. [44]

    Ming Zhou, Jun Luo, Julian Villela, Yaodong Yang, David Rusu, Jiayu Miao, Weinan Zhang, Montgomery Alban, Iman Fadakar, Zheng Chen, Aurora Chongxi Huang, Ying Wen, Kimia Hassanzadeh, Daniel Graves, Dong Chen, Zhengbang Zhu, Nhat M. Nguyen, Mohamed Elsayed, Kun Shao, Sanjeevan ...

    2020 arXiv

  36. [45]

    In International Conference on Machine Learning

    FOP: Factorizing Optimal Joint Policy of Maximum-Entropy Multi-Agent Reinforcement Learning. In International Conference on Machine Learning

  37. [47]

    Yihe Zhou, Shunyu Liu, Yunpeng Qing, Kaixuan Chen, Tongya Zheng, Yanhao Huang, Jie Song, and Mingli Song. 2023. Is Centralized Training with Decentral- ized Execution Framework Centralized Enough for MARL? ArXiv abs/2305.17352 (2023)

    2023 arXiv

  38. [2021]

    ArXiv abs/2110.14300 (2021)

    Multi-Agent Reinforcement Learning for Active Voltage Control on Power Distribution Networks. ArXiv abs/2110.14300 (2021)

    2021 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.