arxiv: 2605.12388 · v2 · submitted 2026-05-12 · 💻 cs.MA · cs.LG

Recognition: unknown

Events as Triggers for Behavioral Diversity in Multi-Agent Reinforcement Learning

Hannes B\"uchi , Manon Flageat , Eduardo Sebasti\'an , Amanda Prorok

Authors on Pith no claims yet

Pith reviewed 2026-05-14 20:37 UTC · model grok-4.3

classification 💻 cs.MA cs.LG

keywords multi-agent reinforcement learningbehavioral diversityeventshypernetworklow-rank adaptationrole reassignmentcooperative tasks

0 comments

The pith

Events trigger dynamic behavior reassignment among agents in multi-agent reinforcement learning while preserving reward maximization.

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

Current multi-agent reinforcement learning methods tie fixed behaviors to fixed agent identities, which prevents effective cooperation when roles must change at precise moments during a task. The paper treats events as observable state changes that mark qualitative shifts in task demands, then builds a continuous manifold of possible behaviors that agents can instantiate on demand. Neural Manifold Diversity supplies a distance measure that works for temporary behaviors independent of any particular agent, while an event-driven hypernetwork produces Low-Rank Adaptation modules to reconfigure a shared policy instantly. The construction is proven to keep the added diversity from reducing the team's reward objective. When this holds, the method becomes the only approach that solves benchmarks requiring sequential behavior reassignment and shows zero-shot generalization to new conditions.

Core claim

The central claim is that events, defined as state changes inducing qualitative task shifts, can decouple agent identity from behavior by letting agents sample from a learned continuous manifold. Neural Manifold Diversity defines a well-posed distance over transient, agent-agnostic behaviors, and an event-based hypernetwork generates Low-Rank Adaptation modules over a shared team policy to enact instant reconfigurations. This arrangement is shown to guarantee that diversity does not interfere with reward maximization, enabling solutions to tasks that demand agents to reassign behaviors in sequence.

What carries the argument

The event-based hypernetwork that generates Low-Rank Adaptation modules over a shared team policy, paired with the Neural Manifold Diversity metric that measures distances between transient, agent-agnostic behaviors.

If this is right

Agents solve tasks that require sequential reassignment of behaviors during execution.
The framework achieves zero-shot generalization to unseen task variations.
Diversity is added without any reduction in the team's reward objective.
The method outperforms established baselines across standard multi-agent cooperation benchmarks.

Where Pith is reading between the lines

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

The same event-triggered manifold could be applied to single-agent settings where objectives change over time.
Automatic inference of events from raw observations would remove the need for explicit state-change detectors.
Storing only the shared policy plus lightweight adaptation modules could reduce memory costs for very large teams.

Load-bearing premise

Events can be detected reliably from observable state changes, and the learned behavior manifold includes every transient action needed to complete the task.

What would settle it

A benchmark task with clear events where the method produces lower team rewards than a fixed-identity baseline because the manifold omits a critical transient behavior sequence.

Figures

Figures reproduced from arXiv: 2605.12388 by Amanda Prorok, Eduardo Sebasti\'an, Hannes B\"uchi, Manon Flageat.

**Figure 1.** Figure 1: Proposed Framework. (a) Rather than shifting behaviors on a fixed timestep or episode schedule, our framework triggers behavioral transitions in response to task events. We realize this through two components: (b) NMD, a diversity metric defined directly on the behavior manifold and independent of which agent executes which behavior; and (c) an event-driven hypernetwork that generates LoRA [Hu et al., 2022… view at source ↗

**Figure 2.** Figure 2: Performance Comparison. Final completion rate, average reward, and episode length across 5 seeds. Colors represent methods: green (Ours), orange (HM: HyperMARL), yellow (CH: CASH), blue (DC: DiCo), purple (PS: Parameter Sharing). Examining each panel reveals how the performance gap scales with task complexity. On Reverse Transport, all methods reach competitive completion rates except PS. The clear separat… view at source ↗

**Figure 3.** Figure 3: Generalization. Final completion rate across 5 seeds for the three generalization axes. White areas denote training distribution; gray areas denote zero-shot evaluation. Colors represent methods: green (Ours), yellow (CH: CASH), blue (DC: DiCo), purple (PS: Parameter Sharing). counterpart to Theorem 3: because diversity is enforced through a projection, the achievable reward does not change as NMDdes is va… view at source ↗

**Figure 4.** Figure 4: Study of Events. (a) Completion rate and (b) average episode reward on Pressure Plate. Colors: green (Ours and SQ: Single Query), orange (HM: HyperMARL), yellow (CH: CASH), blue (DC: DiCo), purple (PS: Parameter Sharing). (c) Agent removal on Pressure Plate and Football. (d) Visualization for four NMD values in Wind Flocking. The arrow indicates wind direction. diversity is supplied to the hypernetwork as … view at source ↗

**Figure 5.** Figure 5: Overview of the six tasks: (a) Dispersion, (b) Navigation, (c) Reverse Transport, (d) [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗

**Figure 6.** Figure 6: Comparison of mean rewards across all evaluation tasks. Learning curves for all the tasks. Colors denote the different methods evaluated: Ours (green), Full Parameter-Sharing (purple), CASH [Fu et al., 2025] (yellow), HyperMARL [Tessera et al., 2025] (orange), and DiCo [Bettini et al., 2024a] (blue). Performance is averaged over 5 seeds [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗

**Figure 7.** Figure 7: LoRA rank ablation in the navigation task. The plot illustrates the mean episode reward [PITH_FULL_IMAGE:figures/full_fig_p020_7.png] view at source ↗

read the original abstract

Effective multi-agent cooperation requires agents to adopt diverse behaviors as task conditions evolve-and to do so at the right moment. Yet, current Multi-Agent Reinforcement Learning (MARL) frameworks that facilitate this diversity are still limited by the fact that they bind fixed behaviors to fixed agent identities. Consequently, they are ill-equipped for tasks where agents need to take on different roles at very specific moments in time. We argue that, to define these behavioral transitions, the missing ingredient is $\textbf{events}$. Events are changes in the state of the system that induce qualitative changes in the task. Based on this view, we introduce a framework that decouples agent identity from behavior, capturing a continuous manifold from which agents instantiate their behaviors in response to events. This framework is based on two elements. First, to build an expressive behavior manifold, we introduce Neural Manifold Diversity (NMD), a formal distance metric that remains well-defined when behaviors are transient and agent-agnostic. Second, we use an event-based hypernetwork that generates Low-Rank Adaptation (LoRA) modules over a shared team policy, enabling on-the-fly agent-policy reconfiguration in response to events. We prove that this construction ensures that diversity does not interfere with reward maximization by design. Empirical results demonstrate that our framework outperforms established baselines across benchmarks while exhibiting zero-shot generalization, and being the only method that solves tasks requiring sequential behavior reassignment.

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.

Desk Editor's Note private letter to a colleague

The paper uses event-triggered hypernetworks to generate LoRA adaptations on a shared policy, paired with a Neural Manifold Diversity metric, to let MARL agents switch behaviors dynamically without fixed identities.

read the letter

The main point is that this work treats events—state changes that shift task requirements—as the signal for agents to reconfigure their policies on the fly. A hypernetwork produces LoRA modules conditioned on those events, while the new Neural Manifold Diversity metric measures spread across transient, agent-agnostic behaviors. The authors claim this construction keeps diversity from trading off against reward by design, and the experiments show gains over baselines plus success on sequential reassignment tasks that others miss, along with zero-shot generalization.

Referee Report

2 major / 1 minor

Summary. The paper introduces a framework for multi-agent reinforcement learning that decouples agent identity from behavior by treating events (state changes inducing qualitative task shifts) as triggers for behavioral transitions. It proposes Neural Manifold Diversity (NMD) as a formal distance metric to capture a continuous, agent-agnostic behavior manifold for transient behaviors, and an event-based hypernetwork that generates LoRA modules over a shared team policy to enable on-the-fly reconfiguration. The authors prove that this construction ensures diversity does not interfere with reward maximization by design, and report empirical outperformance over baselines on benchmarks, zero-shot generalization, and unique success on tasks requiring sequential behavior reassignment.

Significance. If the decoupling proof and empirical results hold, the work could meaningfully advance MARL for dynamic environments where agents must reassign roles at specific moments. Notable strengths include the parameter-free diversity guarantee via the hypernetwork and manifold construction, plus the ability to solve previously intractable sequential reassignment tasks. This addresses a clear limitation in existing identity-bound diversity methods.

major comments (2)

[Proof of decoupling (likely §4)] The central proof that 'this construction ensures that diversity does not interfere with reward maximization by design' (abstract) rests on the assumption that hypernetwork-generated LoRA modules remain decoupled from base-policy gradients under event triggers. The manuscript must specify the exact loss decomposition, gradient constraints, or regularization that prevents the NMD distance term from leaking into shared-policy updates, particularly when events are infrequent or the manifold encodes sequential reassignments.
[Empirical results (likely §5 and Table 2)] The empirical claims of outperformance, zero-shot generalization, and being the only method to solve sequential behavior reassignment tasks require supporting details on baseline implementations, statistical significance testing, and ablations for manifold completeness. Without these, it is difficult to confirm that critical transient behaviors were not omitted during manifold learning.

minor comments (1)

[Methods (event detection)] Clarify how events are reliably detected from raw state changes, including any thresholds or detection mechanisms, to address the weakest assumption noted in the abstract.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. The comments highlight areas where additional clarity and supporting material will strengthen the manuscript. We address each major comment below and will revise accordingly.

read point-by-point responses

Referee: [Proof of decoupling (likely §4)] The central proof that 'this construction ensures that diversity does not interfere with reward maximization by design' (abstract) rests on the assumption that hypernetwork-generated LoRA modules remain decoupled from base-policy gradients under event triggers. The manuscript must specify the exact loss decomposition, gradient constraints, or regularization that prevents the NMD distance term from leaking into shared-policy updates, particularly when events are infrequent or the manifold encodes sequential reassignments.

Authors: We agree that the decoupling argument benefits from an expanded derivation. In the revised §4 we will present the complete loss decomposition: the total objective is L = L_reward + λ L_NMD, where L_NMD is computed exclusively on the hypernetwork outputs (LoRA weights) with an explicit stop-gradient operation applied to the base policy parameters. This ensures gradients from the diversity term never propagate into the shared team policy. The construction remains valid for infrequent events because the manifold is updated continuously offline while event triggers only instantiate the already-decoupled LoRA modules; we will add a short lemma formalizing this separation under sequential reassignments. revision: yes
Referee: [Empirical results (likely §5 and Table 2)] The empirical claims of outperformance, zero-shot generalization, and being the only method to solve sequential behavior reassignment tasks require supporting details on baseline implementations, statistical significance testing, and ablations for manifold completeness. Without these, it is difficult to confirm that critical transient behaviors were not omitted during manifold learning.

Authors: We will augment §5 and the appendix with: (i) precise descriptions and hyperparameter tables for all baselines, (ii) statistical significance results (paired t-tests with p-values and confidence intervals across 10 random seeds), and (iii) an ablation study that varies the number of transient behaviors used to construct the manifold and reports both NMD coverage and task success rates. These additions will demonstrate that the reported performance gains are robust and that the manifold captures the necessary sequential reassignments. revision: yes

Circularity Check

0 steps flagged

No significant circularity; central guarantee is an architectural proof rather than definitional reduction

full rationale

The paper's key claim is a proof that the event-triggered hypernetwork plus NMD manifold 'ensures that diversity does not interfere with reward maximization by design.' This is presented as following from the construction (shared team policy + LoRA modules conditioned on events + NMD distance), not from any equation that defines diversity in terms of reward or vice versa. No fitted parameter is renamed as a prediction, no self-citation supplies a uniqueness theorem, and no ansatz is smuggled. The derivation remains self-contained; empirical outperformance and zero-shot claims are reported separately from the proof. Absent explicit equations showing the diversity term is tautologically equivalent to the reward objective, the load-bearing step does not reduce to its inputs.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 2 invented entities

The framework rests on the assumption that behaviors form a continuous manifold independent of agent identity and that events induce qualitative task changes; new entities NMD and event-hypernetwork are introduced without external validation in the abstract.

free parameters (1)

LoRA adaptation parameters
Rank and scaling factors for the generated modules are trained and thus fitted to data.

axioms (1)

domain assumption Behaviors can be represented on a continuous manifold decoupled from agent identity.
Invoked to justify the NMD metric and hypernetwork design.

invented entities (2)

Neural Manifold Diversity (NMD) no independent evidence
purpose: Formal distance metric for transient, agent-agnostic behaviors.
Newly defined metric central to the diversity component.
Event-based hypernetwork no independent evidence
purpose: Generates LoRA modules on the fly in response to events.
New architectural component for dynamic reconfiguration.

pith-pipeline@v0.9.0 · 5559 in / 1351 out tokens · 34899 ms · 2026-05-14T20:37:12.217458+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages · 2 internal anchors

[1]

Rashid, Tabish and Samvelyan, Mikayel and De Witt, Christian Schroeder and Farquhar, Gregory and Foerster, Jakob and Whiteson, Shimon , journal=

work page
[2]

International Conference on Machine Learning , year=

Controlling Behavioral Diversity in Multi-Agent Reinforcement Learning , author=. International Conference on Machine Learning , year=

work page
[3]

Conference on Robot Learning , pages=

Capability-Aware Shared Hypernetworks for Flexible Heterogeneous Multi-Robot Coordination , author=. Conference on Robot Learning , pages=. 2025 , organization=

work page 2025
[4]

Journal of Machine Learning Research , year =

Matteo Bettini and Ajay Shankar and Amanda Prorok , title =. Journal of Machine Learning Research , year =

work page
[5]

2023 , booktitle =

Heterogeneous Multi-Robot Reinforcement Learning , author =. 2023 , booktitle =

work page 2023
[6]

2024 , publisher=

Multi-agent reinforcement learning: Foundations and modern approaches , author=. 2024 , publisher=

work page 2024
[7]

Tessera, Kale-ab Abebe and Rahman, Arrasy and Storkey, Amos and Albrecht, Stefano V , booktitle=

work page
[8]

Artificial Intelligence Review , volume=

Multi-agent deep reinforcement learning: a survey , author=. Artificial Intelligence Review , volume=. 2022 , publisher=

work page 2022
[9]

Advances in Neural Information Processing Systems , volume=

Celebrating diversity in shared multi-agent reinforcement learning , author=. Advances in Neural Information Processing Systems , volume=

work page
[10]

HyperNetworks

Hypernetworks , author=. arXiv preprint arXiv:1609.09106 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[11]

IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews) , volume=

A comprehensive survey of multiagent reinforcement learning , author=. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews) , volume=. 2008 , publisher=

work page 2008
[12]

International Symposium on Distributed Autonomous Robotic Systems , organization=

Bettini, Matteo and Kortvelesy, Ryan and Blumenkamp, Jan and Prorok, Amanda , year =. International Symposium on Distributed Autonomous Robotic Systems , organization=

work page
[13]

Proceedings of the 22nd International Conference on Autonomous Agents and Multiagent Systems , publisher =

Bettini, Matteo and Shankar, Ajay and Prorok, Amanda , year =. Proceedings of the 22nd International Conference on Autonomous Agents and Multiagent Systems , publisher =

work page
[14]

, journal=

Parker, L.E. , journal=. 1998 , volume=

work page 1998
[15]

arXiv preprint arXiv:2412.16244 , year=

The impact of behavioral diversity in multi-agent reinforcement learning , author=. arXiv preprint arXiv:2412.16244 , year=

work page arXiv
[16]

Wang, Tonghan and Dong, Heng and Lesser, Victor and Zhang, Chongjie , booktitle=

work page
[17]

ACM Transactions on Intelligent Systems and Technology , volume=

ROIS: Role-Based Multi-Agent Collaboration by Context-Time-Aware Information Sharing , author=. ACM Transactions on Intelligent Systems and Technology , volume=. 2026 , publisher=

work page 2026
[18]

2018 , author =

Proceedings of the 35th International Conference on Machine Learning , url =. 2018 , author =

work page 2018
[19]

Neurocomputing , volume=

Heterogeneous multi-agent reinforcement learning for zero-shot scalable collaboration , author=. Neurocomputing , volume=. 2025 , publisher=

work page 2025
[20]

Li, Xinran and Pan, Ling and Zhang, Jun , journal=

work page
[21]

Liang, Yongyuan and Xu, Tingqiang and Hu, Kaizhe and Jiang, Guangqi and Huang, Furong and Xu, Huazhe , journal=

work page
[22]

Xiong, Zheng and Li, Kang and Wang, Zilin and Jackson, Matthew and Foerster, Jakob and Whiteson, Shimon , journal=

work page
[23]

Hegde, Shashank and Das, Satyajeet and Salhotra, Gautam and Sukhatme, Gaurav S , booktitle=

work page
[24]

arXiv preprint arXiv:2506.09434 , year=

When is diversity rewarded in cooperative multi-agent learning? , author=. arXiv preprint arXiv:2506.09434 , year=

work page arXiv
[25]

International Conference on Autonomous Agents and Multiagent Systems , pages=

Parameter Sharing with Network Pruning for Scalable Multi-Agent Deep Reinforcement Learning , author=. International Conference on Autonomous Agents and Multiagent Systems , pages=

work page
[26]

2305.17330 , archivePrefix=

Zhengbang Zhu and Minghuan Liu and Liyuan Mao and Bingyi Kang and Minkai Xu and Yong Yu and Stefano Ermon and Weinan Zhang , year=. 2305.17330 , archivePrefix=

work page arXiv
[27]

Hu, Edward J and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Liang and Chen, Weizhu and others , journal=

work page
[28]

2502.05573 , archivePrefix=

Beining Zhang and Aditya Kapoor and Mingfei Sun , year=. 2502.05573 , archivePrefix=

work page arXiv
[29]

arXiv preprint arXiv:2304.07645 , year=

Magnitude invariant parametrizations improve hypernetwork learning , author=. arXiv preprint arXiv:2304.07645 , year=

work page arXiv
[30]

International Conference on Machine Learning , pages=

Scaling multi-agent reinforcement learning with selective parameter sharing , author=. International Conference on Machine Learning , pages=. 2021 , organization=

work page 2021
[31]

2211.15457 , archivePrefix=

Sahand Rezaei-Shoshtari and Charlotte Morissette and Francois Robert Hogan and Gregory Dudek and David Meger , year=. 2211.15457 , archivePrefix=

work page arXiv
[32]

Chauhan, Vinod Kumar and Zhou, Jiandong and Lu, Ping and Molaei, Soheila and Clifton, David A. , year=. A brief review of hypernetworks in deep learning , volume=. Artificial Intelligence Review , publisher=. doi:10.1007/s10462-024-10862-8 , number=

work page doi:10.1007/s10462-024-10862-8
[33]

Advances in neural information processing systems , volume=

On the modularity of hypernetworks , author=. Advances in neural information processing systems , volume=

work page
[34]

2510.02630 , archivePrefix=

Hao Zhang and Zhenjia Li and Runfeng Bao and Yifan Gao and Xi Xiao and Bo Huang and Yuhang Wu and Tianyang Wang and Hao Xu , year=. 2510.02630 , archivePrefix=

work page arXiv
[35]

arXiv preprint arXiv:2312.08399 , year=

Principled weight initialization for hypernetworks , author=. arXiv preprint arXiv:2312.08399 , year=

work page arXiv
[36]

2502.12605 , archivePrefix=

Kyeonghyeon Park and David Molina Concha and Hyun-Rok Lee and Chi-Guhn Lee and Taesik Lee , year=. 2502.12605 , archivePrefix=

work page arXiv
[37]

Science Robotics , volume =

Amanda Prorok , title =. Science Robotics , volume =. 2025 , doi =. https://www.science.org/doi/pdf/10.1126/scirobotics.adv4049 , abstract =

work page doi:10.1126/scirobotics.adv4049 2025
[38]

Revisiting parameter sharing in multi-agent deep reinforcement learning,

Revisiting parameter sharing in multi-agent deep reinforcement learning , author=. arXiv preprint arXiv:2005.13625 , year=

work page arXiv 2005
[39]

Proceedings of the IEEE international conference on computer vision , pages=

Delving deep into rectifiers: Surpassing human-level performance on imagenet classification , author=. Proceedings of the IEEE international conference on computer vision , pages=

work page
[40]

2010 , editor =

Glorot, Xavier and Bengio, Yoshua , booktitle =. 2010 , editor =

work page 2010
[41]

L. S. Shapley , title =. Proceedings of the National Academy of Sciences , volume =

work page
[42]

Forty-first International Conference on Machine Learning , year=

Dora: Weight-decomposed low-rank adaptation , author=. Forty-first International Conference on Machine Learning , year=

work page
[43]

Wang, Tonghan and Gupta, Tarun and Mahajan, Anuj and Peng, Bei and Whiteson, Shimon and Zhang, Chongjie , booktitle=

work page
[44]

International Conference on Autonomous Agents and Multiagent Systems , pages=

Cooperative multi-agent control using deep reinforcement learning , author=. International Conference on Autonomous Agents and Multiagent Systems , pages=. 2017 , organization=

work page 2017
[45]

2009 , publisher=

Optimal transport: old and new , author=. 2009 , publisher=

work page 2009
[46]

Linear Algebra and its Applications , volume=

The distance between two random vectors with given dispersion matrices , author=. Linear Algebra and its Applications , volume=. 1982 , publisher=

work page 1982
[47]

2025 , organization=

Goel, Harsh and Omama, Mohammad and Chalaki, Behdad and Tadiparthi, Vaishnav and Pari, Ehsan Moradi and Chinchali, Sandeep P , booktitle=. 2025 , organization=

work page 2025
[48]

2025 , volume=

Liu, Yirui , booktitle=. 2025 , volume=

work page 2025
[49]

Agent oriented software engineering with

Pav. Agent oriented software engineering with. International Central and Eastern European Conference on Multi-Agent Systems , pages=. 2003 , organization=

work page 2003
[50]

A holonic approach to

Cossentino, Massimo and Gaglio, Salvatore and Garro, Alfredo and Russo, Wilma , booktitle=. A holonic approach to. 2005 , organization=

work page 2005
[51]

Agent-Oriented Software Engineering XI , pages=

Agent oriented software engineering , author=. Agent-Oriented Software Engineering XI , pages=. 2011 , organization=

work page 2011
[52]

2014 , publisher=

Agent-oriented design processes , author=. 2014 , publisher=

work page 2014
[53]

PyTorch: An Imperative Style, High-Performance Deep Learning Library

Adam Paszke and Sam Gross and Francisco Massa and Adam Lerer and James Bradbury and Gregory Chanan and Trevor Killeen and Zeming Lin and Natalia Gimelshein and Luca Antiga and Alban Desmaison and Andreas Köpf and Edward Yang and Zach DeVito and Martin Raison and Alykhan Tejani and Sasank Chilamkurthy and Benoit Steiner and Lu Fang and Junjie Bai and Soumi...

work page internal anchor Pith review Pith/arXiv arXiv 1912
[54]

James Bradbury and Roy Frostig and Peter Hawkins and Matthew James Johnson and Chris Leary and Dougal Maclaurin and George Necula and Adam Paszke and Jake Vander

work page
[55]

Jonathan Heek and Anselm Levskaya and Avital Oliver and Marvin Ritter and Bertrand Rondepierre and Andreas Steiner and Marc van

work page
[56]

International Conference on Machine Learning , pages=

The emergence of individuality , author=. International Conference on Machine Learning , pages=. 2021 , organization=

work page 2021
[57]

IEEE International Conference on Acoustics, Speech and Signal Processing , pages=

Adaptive parameter sharing for multi-agent reinforcement learning , author=. IEEE International Conference on Acoustics, Speech and Signal Processing , pages=

work page
[58]

IEEE Conference on Games , pages=

Policy diversity for cooperative agents , author=. IEEE Conference on Games , pages=

work page
[59]

Advances in Neural Information Processing Systems , volume=

Maven: Multi-agent variational exploration , author=. Advances in Neural Information Processing Systems , volume=

work page
[60]

International Conference on Swarm Intelligence , pages=

Natural emergence of heterogeneous strategies in artificially intelligent competitive teams , author=. International Conference on Swarm Intelligence , pages=. 2021 , organization=

work page 2021
[61]

AAAI Conference on Artificial Intelligence , volume=

Google research football: A novel reinforcement learning environment , author=. AAAI Conference on Artificial Intelligence , volume=

work page