Pith. sign in

REVIEW 5 major objections 4 minor 18 references

Wasserstein-Barycenter Consensus for Cooperative Multi-Agent Reinforcement Learning

T0 review · 5 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Cooperative MARL agents can be aligned by penalizing each policy's Sinkhorn distance to a team Wasserstein barycenter, with geometric convergence to a common policy.

desk verdict Promising barycenter-consensus idea for cooperative MARL, but the central convergence proof is an assertion and the experiment is too thin to carry the claims. read the letter →

arxiv 2506.12497 v2 pith:JE5LMHLA submitted 2025-06-14 eess.SY cs.SY

Ali Baheri This is my paper · ORCID
classification eess.SYcs.SY
keywords multi-agentreinforcementlearningcooperativeMARLoptimaltransportWassersteinbarycenterSinkhorndivergencepolicyconsensusgeometricconvergencegradient
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

This paper introduces a consensus mechanism for cooperative multi-agent reinforcement learning based on optimal transport. The team's collective strategy is defined as the entropic-regularized $p$-Wasserstein barycenter of the agents' joint state–action visitation measures, and each agent's policy objective is augmented with a soft penalty proportional to its Sinkhorn divergence from this barycenter. The paper proves that, under compactness and Lipschitz assumptions, the maximal pairwise Sinkhorn divergence between agents contracts at a geometric rate, so all policies converge to a common distribution. A cooperative navigation experiment shows the method reaches higher final reward and converges faster than independent PPO, KL-regularized, and parameter-sharing baselines.

What carries the argument

The central object is the entropic-regularized $p$-Wasserstein barycenter, the measure $\mu^{*(t)}$ minimizing $\frac{1}{N}\sum_i W_{p,\varepsilon}^p(\mu, \mu_i^{(t)}) + \varepsilon\, \mathrm{KL}(\mu \| \eta)$, computed by alternating Sinkhorn (iterative Bregman projection) updates on the discretized joint state–action space. The mechanism that couples consensus to learning is the gradient identity $\nabla_{\theta_i} W_{p,\varepsilon}^p(\mu_i^{(t)}, \mu^{*(t)}) = \int d(x,y)^p\, \nabla_{\theta_i}\log\pi_i(a|s;\theta_i)\, d\gamma_i^{*(t)}(x,y)$, where $\gamma_i^{*(t)}$ is the optimal coupling between $\mu_i^{(t)}$ and the barycenter; this makes the entire loop differentiable by sampling from the visitation measure. The convergence proof uses the contraction of a single gradient step toward the barycenter (attributed to strong convexity of the regularized OT problem and Lipschitz continuity of the policy-gradient mapping), combined with the triangle inequality for $W_{p,\varepsilon}$ and an induction over iterations.

What would settle it

Run the WBC algorithm on a two-agent grid task and record $\log D(t)$ over iterations; a lack of linear decay (i.e., $D(t+1)/D(t)$ not consistently below 1) would contradict the geometric-rate claim. Alternatively, compute the constant $C$ explicitly for a Gaussians-with-known-barycenter instance and check whether $1-\alpha\lambda C < 1$ under the paper's stated assumptions.

Watch

Extended reading notes

Core claim

The central claim is that consensus in cooperative MARL can be achieved geometrically: rather than sharing parameters or a centralized critic, agents are drawn together by a soft penalty equal to their entropic-regularized $p$-Wasserstein distance to a common barycenter. The paper's main theoretical result (Theorem 3.1) states that if the state–action space is compact, rewards are bounded and Lipschitz, policy-gradient operators are $L$-Lipschitz in the induced distributions, and the step size and consensus weight satisfy $\kappa = 1 - \alpha\lambda C < 1$ for a constant $C > 0$, then the maximal pairwise discrepancy $D(t) = \max_{i,j} W_{p,\varepsilon}(\mu_i^{(t)}, \mu_j^{(t)})$ obeys $D(t+1) \leq \kappa D(t)$ and hence converges to zero at a geometric rate. The paper also proves a fast-rate bound (Theorem 3.2) on the accuracy of the estimated barycenter as a function of sample size. Together these results position the Wasserstein barycenter as a principled, differentiable consensus anchor that preserves specialized behavior while aligning the team.

Load-bearing premise

The geometric convergence proof depends on an unproved contraction step: a single gradient step with weight $\alpha\lambda$ is asserted to shrink each agent's Sinkhorn distance to the current barycenter by a factor $(1-\alpha\lambda C)$, where $C$ is never constructed and the reward-gradient term is ignored.

Editorial extensions

If this is right

  • Coordination no longer requires parameter sharing, centralized critics, or hand-designed reward shaping; a soft OT penalty provides a differentiable, geometry-aware alignment signal.
  • The theorem yields a finite-time bound on policy disagreement: after $t$ iterations, the maximal pairwise Sinkhorn divergence is at most $\kappa^t D(0)$, so the time to reach a desired consensus tolerance is $\mathcal{O}(\log(1/\delta))$.
  • The fast-rate barycenter estimation result implies that, under strong convexity and bounded density, only $m = O(\eta^{-1} \log(N/\delta))$ samples per agent are needed to achieve barycenter error $\eta$, making the consensus step statistically efficient.
  • The empirical comparison suggests the same mechanism is competitive with or better than existing consensus regularizers: about 2× the final reward of independent PPO and roughly 50% higher than a KL-regularized baseline on the cooperative navigation task.

Reading between the lines

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

  • A constructive proof that makes the contraction constant $C$ explicit in terms of $\varepsilon$, the Lipschitz constants, and the strong-convexity parameter would turn Theorem 3.1 into a practical recipe for choosing $\alpha$ and $\lambda$, a step the paper leaves open.
  • The same barycenter-consensus penalty could plausibly be applied to critics or value distributions rather than visitation measures; whether the geometric-rate guarantee carries over to off-policy settings is a direct testable extension.
  • The predicted geometric decay of pairwise Sinkhorn divergence is measurable in training logs: if the log-ratio $D(t+1)/D(t)$ is not eventually bounded below 1 on a standard benchmark, the claimed contraction rate would fail, even if the method still performs well.
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

5 major / 4 minor

Summary. The paper proposes a cooperative MARL algorithm in which the team's consensus policy is the entropic-regularized p-Wasserstein barycenter of the agents' empirical visitation measures. Each agent maximizes the expected team reward minus a Sinkhorn-divergence penalty to this barycenter, and the paper claims in Theorem 3.1 that the maximal pairwise policy discrepancy contracts geometrically under Lipschitz and compactness assumptions. Theorem 3.2 claims a fast statistical rate for the Sinkhorn barycenter. A small cooperative-navigation experiment compares the method with independent PPO, KL-regularized, and parameter-sharing baselines. The reviewer finds that the proof of Theorem 3.1 does not establish the central contraction claim: the key inequality is asserted without proof, the reward-gradient term is omitted, the contraction constant is never constructed, and the proof uses a different contraction factor from the theorem statement. The appendix proof of Theorem 3.2 also contains unresolved inconsistencies. The empirical results are transparent but preliminary.

Significance. The high-level idea of using an OT barycenter as a soft consensus target rather than hard parameter sharing is appealing and worth exploring. The paper is clearly structured, the algorithm is specified in detail, and the inclusion of a limitations section is commendable. However, the main advertised contribution is the geometric convergence theorem, and that contribution is not supported by the provided argument. The experimental evaluation is a proof-of-concept on a single small task without error bars. If a rigorous proof could be supplied under explicit, verifiable assumptions, this framework would be a useful contribution; in its current form, the paper does not meet the standard for publication.

major comments (5)
  1. [§3, Theorem 3.1 proof] The load-bearing inequality W_{p,ε}(μ_i^{(t+1)}, μ^{*(t)}) ≤ (1 − αλC) W_{p,ε}(μ_i^{(t)}, μ^{*(t)}) is asserted without proof. The update is a parameter-space gradient step, and μ_i^{(t+1)} is the empirical visitation distribution after executing the updated policy; neither the strong convexity of the entropic OT problem nor L-Lipschitz continuity of the policy-gradient mapping, as stated, implies a contraction of this quantity. The reward-gradient term +α∇_{θ_i} E[R] is absent from the inequality, and no assumption in the theorem controls its effect on the Sinkhorn distance to the barycenter. Without a derivation of this inequality, the conclusion D^{(t+1)} ≤ κ D^{(t)} does not follow.
  2. [§3, Theorem 3.1] The theorem statement imposes κ = 1 − αλC < 1, but the proof sets κ = 2(1 − αλC) < 1, which is a strictly stronger condition requiring αλC > 1/2. This stronger condition is not stated, and the constant C is never constructed or bounded in terms of the entropic regularizer and Lipschitz constants. As a result, the hypothesis of the theorem is not verifiable from the stated assumptions.
  3. [§3, Theorem 3.1 proof] The proof invokes the triangle inequality for W_{p,ε}. The entropic-regularized Sinkhorn divergence is not a metric, and the triangle inequality is not a general property of this divergence for ε > 0. If the authors intend to use a distance that satisfies the triangle inequality, they must state and prove that property for their W_{p,ε}; otherwise the step bounding W_{p,ε}(μ_i^{(t+1)}, μ_j^{(t+1)}) by the sum of two distances to μ^{*(t)} is unjustified.
  4. [§3, gradient of the consensus penalty] The displayed formula ∇_{θ_i} W^p_{p,ε}(μ_i^{(t)}, μ^{*(t)}) = ∫ d(x,y)^p ∇_{θ_i} log π_i(a|s;θ_i) dγ_i^{*(t)}(x,y) is not justified as written, because μ_i^{(t)} is the empirical visitation distribution induced by the current policy and therefore depends on θ_i. Differentiating only the action probabilities while holding the state-action measure fixed omits the state-distribution term, so the update implemented in Algorithm 1 is not demonstrably the update analyzed in Theorem 3.1.
  5. [Appendix A.1, proof of Theorem 3.2] The proof sketch and the appendix give inconsistent bounded-difference bounds: the sketch states D^p/(N m ε_m) = O(m^{-2}), while (A.1) uses D^p/(N m), a factor of ε_m = D^p/m apart. Step 2's metric-entropy bound log M_γ ≤ C_d γ^{-d} for P(X) is asserted without reference to the dimension or structure of X, and Step 3's inequality F(μ) ≥ F(μ*) + λ_min W_{p,ε}(μ, μ*) does not follow from λ_min-strong convexity in any stated norm. The claimed rate in Eq. (1) is therefore not established; the limitation section's concession that the strong-convexity and density-boundedness assumptions seldom hold does not repair this gap.
minor comments (4)
  1. [§2, §3, notation] The paper alternates between W^p_{p,ε} in the objectives and W_{p,ε} in Theorem 3.1 and in the definition of D^{(t)}; since contraction in W_{p,ε} does not automatically imply contraction in W^p_{p,ε}, the notation and the logical relation between the two should be clarified.
  2. [§3, Theorem 3.1 assumptions] The assumption that the policy-gradient operators are L-Lipschitz 'in the induced distributions' is not defined precisely; the objects, norms, and the space on which the Lipschitz condition is imposed should be specified.
  3. [§4, Figure 3] Figure 3 reports point estimates without error bars, number of seeds, or training details, so the claim of approximately 2× improvement over IPPO is not substantiated beyond a single run.
  4. [§3, barycenter computation] The iterative update for v^{(ℓ+1)} appears to be a simplified fixed-point update, whereas standard Sinkhorn-barycenter algorithms maintain per-measure dual potentials; this step should be clarified or referenced.

Circularity Check

1 steps flagged · score 8.0 of 10

Theorem 3.1's geometric contraction is asserted as an existence assumption, making the consensus convergence result a restatement of the proof's unproved contraction step.

  1. self definitional [Theorem 3.1 (Convergence to Consensus), Section 3, proof]
    "By the strong convexity of the entropic-regularized OT problem and the Lipschitz continuity of the policy-gradient mapping, there exists C > 0 such that a gradient step with weight αλ contracts each agent's distance to the barycenter: W_{p,ε}(μ_i^{(t+1)}, μ^{*(t)}) ≤ (1 − α λ C) W_{p,ε}(μ_i^{(t)}, μ^{*(t)}). ... Setting κ = 2(1 − αλC) < 1 yields max_{i,j} W_{p,ε}(μ_i^{(t+1)}, μ_j^{(t+1)}) ≤ κ D(t), and by induction D(t) ≤ κ^t D(0) → 0."

    The load-bearing contraction inequality is not derived from the stated Lipschitz, boundedness, or compactness assumptions; it is posited as 'there exists C > 0' directly. The theorem's stated hypothesis κ = 1 − αλC < 1 is the same contraction condition rewritten, and since C is never constructed or bounded, the hypothesis is a restatement of the conclusion. The proof then combines this assumed per-agent contraction with the triangle inequality to obtain pairwise contraction, so the geometric convergence D(t+1) ≤ κD(t) reduces to the unproved existence of C.

full rationale

The central theoretical claim of the paper is Theorem 3.1, which states that the maximal pairwise Sinkhorn divergence contracts geometrically. Its proof rests entirely on the assertion that 'there exists C > 0' such that each agent's distance to the barycenter contracts by the factor (1 − αλC). This is exactly the geometric contraction the theorem claims to prove; the constant C is never constructed, bounded, or related to the reward gradient, so the theorem's hypothesis κ = 1 − αλC < 1 is a restatement of the contraction rather than an independent condition. The subsequent triangle-inequality step and induction do not add substance: they convert the assumed per-agent contraction into pairwise contraction. The proof also changes the contraction factor from the theorem's 1−αλC to 2(1−αλC)<1, imposing the unstated condition αλC > 1/2. The paper's own Limitations paragraph concedes that the fast-rate theorem's assumptions 'seldom hold in high-dimensional continuous control tasks,' but this concerns Theorem 3.2; the more fundamental gap is the assumed contraction in Theorem 3.1. Self-citations (e.g., Baheri 2023; Baheri et al. 2025; Shahrooei & Baheri 2025) appear only as contextual references and are not load-bearing. The Sinkhorn-barycenter convergence and the fast-rate result in Theorem 3.2 draw on external results such as Genevay et al. and Villani, which are independent evidence. Thus the circularity is localized to the central convergence proof: the main theorem's conclusion is effectively assumed through the existence of C, making the derivation circular rather than merely incomplete.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central theorem depends on an unproved contraction constant C and strong-convexity assumptions that the paper itself calls restrictive. The experimental hyperparameters are hand-chosen without sensitivity analysis. No new physical or mathematical entities are introduced.

free parameters (5)
  • λ (consensus weight) = 0.5
    Controls the strength of the Sinkhorn-divergence penalty; the paper provides no sensitivity analysis.
  • ε (entropic regularization) = 0.1
    Chosen by hand for the experiment; a decreasing schedule is mentioned but not specified numerically.
  • β (ground metric balance) = 0.8
    Balances state and action distances in the ground metric d((s,a),(s',a')).
  • C (contraction constant in Theorem 3.1) = unspecified
    Asserted to exist so that κ = 1 − αλC < 1; never constructed or bounded, effectively a free parameter chosen to force the contraction result.
  • Adaptive schedules for λ and ε = qualitative
    Described as warm-starting with large λ and annealing ε downward, but no concrete schedule or values are given.
assumptions (6)
  • domain assumption State-action space X is compact and endowed with a ground metric d((s,a),(s',a')) = ||s-s'|| + β||a-a'||.
    Used in Theorem 3.1 to guarantee the barycenter exists; standard in theoretical RL but restrictive for large or continuous state-action spaces.
  • domain assumption Rewards are bounded and Lipschitz; policy-gradient operators are L-Lipschitz in the induced distributions.
    Stated in Theorem 3.1 as needed for the contraction constant C, but no proof is given that the parameterized policies satisfy these conditions.
  • domain assumption The entropic-regularized barycenter functional μ ↦ (1/N) Σ_i W_{p,ε}(μ, μ_i) is λmin-strongly convex.
    Assumed in Theorem 3.2 to convert uniform deviation bounds into a bound on the barycenter itself; the paper's Limitations section concedes this rarely holds in high-dimensional continuous control.
  • domain assumption Data-generating measures admit densities with 0 < ρ ≤ ρ_i ≤ ρ̄ < ∞ and X has finite dimension d.
    Assumed in Theorem 3.2 for the bounded-difference and metric-entropy steps.
  • standard math Sinkhorn divergence is Lipschitz in its arguments with modulus D^p.
    Invoked in Appendix A.1 via Lemma 2.1 of Genevay et al. (2018) to control the effect of replacing one sample.
  • standard math Metric-entropy bound log M_γ ≤ C_d γ^{-d} for probability measures on a compact d-dimensional space.
    Cited from Villani et al. (2009) and used for the uniform-concentration argument; implicitly assumes finite dimension d.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Wasserstein-Barycenter Consensus for Cooperative Multi-Agent Reinforcement Learning." pith.science (2026). https://pith.science/paper/JE5LMHLA

@misc{pith2026250612497,
  author       = {Pith},
  title        = {Pith review of: Wasserstein-Barycenter Consensus for Cooperative Multi-Agent Reinforcement Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JE5LMHLA}},
  note         = {Machine review of arXiv:2506.12497}
}
abstract

Cooperative multi-agent reinforcement learning (MARL) demands principled mechanisms to align heterogeneous policies while preserving the capacity for specialized behavior. We introduce a novel consensus framework that defines the team strategy as the entropic-regularized $p$-Wasserstein barycenter of agents' joint state--action visitation measures. By augmenting each agent's policy objective with a soft penalty proportional to its Sinkhorn divergence from this barycenter, the proposed approach encourages coherent group behavior without enforcing rigid parameter sharing. We derive an algorithm that alternates between Sinkhorn-barycenter computation and policy-gradient updates, and we prove that, under standard Lipschitz and compactness assumptions, the maximal pairwise policy discrepancy contracts at a geometric rate. Empirical evaluation on a cooperative navigation case study demonstrates that our OT-barycenter consensus outperforms an independent learners baseline in convergence speed and final coordination success.

Figures

Figures reproduced from arXiv: 2506.12497 by the authors.

Figure 1
Figure 1. Policy Action Maps comparing Baseline (Independent) Policy with Wasserstein-Barycenter Consensus Policy. Color cod￾ing indicates different actions: red (right), blue (left), green (up), orange (down), and gray (stay). Stars indicate target positions. Let µ ⋆ = arg minµ 1 N PN i=1 Wp,εm(µ, µi) and µˆ ⋆ = arg minµ 1 N PN i=1 Wp,εm(µ, µˆi). Then for every δ ∈ (0, 1), with probability at least 1 − δ, Wp,εm [PITH_FULL_I… view at source ↗
Figure 2
Figure 2. Comparison of action probability distributions between baseline (Independent PPO) agents (top) and Wasserstein-Barycenter Consensus (WBC) agents (bottom) across five representative states. Cell values indicate the probability of selecting each action in the given state, with higher probabilities shown in brighter colors. 0 25 50 75 100 125 150 175 200 Training Iteration 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Average Te… view at source ↗
Figure 3
Figure 3. Average team reward over training iterations comparing WBC against baseline methods. WBC achieves superior convergence speed and final performance. action (0.51, 0.56, and 0.60, respectively), while for State 2, they all prioritize the “left” action. This alignment confirms the theoretical guarantee of policy convergence provided by our consensus mechanism [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

18 extracted references · 11 canonical work pages

  1. [1]

    Understanding reward ambiguity through optimal transport theory in inverse reinforcement learning

    Baheri, A. Understanding reward ambiguity through optimal transport theory in inverse reinforcement learning. arXiv preprint arXiv:2310.12055, 2023

    arXiv 2023

  2. [2]

    W AVE : Wasserstein adaptive value estimation for actor-critic reinforcement learning

    Baheri, A., Sharooei, Z., and Salgarkar, C. W AVE : Wasserstein adaptive value estimation for actor-critic reinforcement learning. Proceedings of Machine Learning Research vol, 283: 0 1--12, 2025

    work page 2025

  3. [3]

    A comprehensive survey of multiagent reinforcement learning

    Busoniu, L., Babuska, R., and De Schutter, B. A comprehensive survey of multiagent reinforcement learning. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), 38 0 (2): 0 156--172, 2008

    work page 2008

  4. [4]

    Sinkhorn distances: Lightspeed computation of optimal transport

    Cuturi, M. Sinkhorn distances: Lightspeed computation of optimal transport. Advances in neural information processing systems, 26, 2013

    2013

  5. [5]

    Counterfactual multi-agent policy gradients

    Foerster, J., Farquhar, G., Afouras, T., Nardelli, N., and Whiteson, S. Counterfactual multi-agent policy gradients. In Proceedings of the AAAI conference on artificial intelligence, volume 32, 2018

    work page 2018

  6. [6]

    Learning generative models with sinkhorn divergences

    Genevay, A., Peyr \'e , G., and Cuturi, M. Learning generative models with sinkhorn divergences. In International Conference on Artificial Intelligence and Statistics, pp.\ 1608--1617. PMLR, 2018

    work page 2018

  7. [7]

    On centralized critics in multi-agent reinforcement learning

    Lyu, X., Baisero, A., Xiao, Y., Daley, B., and Amato, C. On centralized critics in multi-agent reinforcement learning. Journal of Artificial Intelligence Research, 77: 0 295--354, 2023

    work page 2023

  8. [8]

    Offline reinforcement learning with wasserstein regularization via optimal transport maps

    Omura, M., Mukuta, Y., Ota, K., Osa, T., and Harada, T. Offline reinforcement learning with wasserstein regularization via optimal transport maps. In Reinforcement Learning Conference

    work page

Show all 18 references
  1. [9]

    Computational optimal transport: With applications to data science

    Peyr \'e , G., Cuturi, M., et al. Computational optimal transport: With applications to data science. Foundations and Trends in Machine Learning , 11 0 (5-6): 0 355--607, 2019

    2019

  2. [10]

    and Baheri, A

    Shahrooei, Z. and Baheri, A. Risk-averse reinforcement learning: An optimal transport perspective on temporal difference learning. arXiv preprint arXiv:2502.16328, 2025

    2025 arXiv

  3. [11]

    and Lu, Z

    Su, K. and Lu, Z. Divergence-regularized multi-agent actor-critic. In International Conference on Machine Learning, pp.\ 20580--20603. PMLR, 2022

    2022

  4. [12]

    K., Grammel, N., Son, S., Black, B., and Agrawal, A

    Terry, J. K., Grammel, N., Son, S., Black, B., and Agrawal, A. Revisiting parameter sharing in multi-agent deep reinforcement learning. arXiv preprint arXiv:2005.13625, 2020

    2005 arXiv

  5. [13]

    Villani, C. et al. Optimal transport: old and new, volume 338. Springer, 2009

    2009

  6. [14]

    Reaching consensus in cooperative multi-agent reinforcement learning with goal imagination

    Wang, L., Zhu, K., Zhu, F., Yao, X., Zhang, S., Ye, D., Fu, H., Fu, Q., and Yang, W. Reaching consensus in cooperative multi-agent reinforcement learning with goal imagination. arXiv preprint arXiv:2403.03172, 2024

    2024 arXiv

  7. [15]

    Learning to share in networked multi-agent reinforcement learning

    Yi, Y., Li, G., Wang, Y., and Lu, Z. Learning to share in networked multi-agent reinforcement learning. Advances in Neural Information Processing Systems, 35: 0 15119--15131, 2022

    2022

  8. [16]

    A survey of progress on cooperative multi-agent reinforcement learning in open environment

    Yuan, L., Zhang, Z., Li, L., Guan, C., and Yu, Y. A survey of progress on cooperative multi-agent reinforcement learning in open environment. arXiv preprint arXiv:2312.01058, 2023

    2023 arXiv

  9. [17]

    A semi-independent policies training method with shared representation for heterogeneous multi-agents reinforcement learning

    Zhao, B., Jin, W., Chen, Z., and Guo, Y. A semi-independent policies training method with shared representation for heterogeneous multi-agents reinforcement learning. Frontiers in Neuroscience, 17: 0 1201370, 2023

    2023

  10. [18]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.