Pith. sign in

REVIEW 1 major objections 6 minor 66 references

Heterogeneous Multi-agent Multi-armed Bandits on Stochastic Block Models

T0 review · 1 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read A cooperative bandit algorithm on stochastic block models achieves instance-dependent regret of order (C/M) log T, where C is the number of clusters and M the number of agents.

desk verdict Fresh problem formulation with a plausible C/M regret improvement, but the main relaxed-edge-probability theorem has a correctable exponent error in its connectivity proof. read the letter →

arxiv 2502.08003 v1 pith:WTS5TU7C submitted 2025-02-11 cs.LG

Mengfan Xu , Liren Shan , Fatemeh Ghaffari , Xuchuang Wang , Xutong Liu , Mohammad Hajiesmaili This is my paper · ORCID
classification cs.LG MSC 68Q3262L0505C80
keywords multi-agentmulti-armedbanditsstochasticblockmodelsheterogeneousrewardsregretboundsrandomcommunicationgraphsconsensusclusterdetectionUCB
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 studies multi-agent multi-armed bandits in which both the communication graph and the reward distributions inherit a cluster structure: agents within a cluster share the same reward means and denser edges, while agents in different clusters differ. The authors propose an algorithm, UCB-SBM, that aggregates rewards at the cluster level and exchanges only cluster-level statistics, and they prove that its regret is O((C/M) log T), where C is the number of clusters and M the number of agents, under sub-Gaussian rewards. This is the first bound to exhibit dependence on the degree of heterogeneity, C/M, interpolating between the fully homogeneous case (C=1) and the fully heterogeneous case (C=M), and it improves on the prior O($M^{2}$ log T) bound while requiring much weaker lower bounds on edge probabilities. The same guarantee holds when the cluster assignment is unknown, at the price of a cluster-detection step and an extra additive O(M) burn-in cost.

What carries the argument

The load-bearing object is the cluster-level subgraph G^C_t, whose vertices are clusters and whose edges indicate whether any agent pair across two clusters communicated at time t. Lemma 6 shows the cluster-edge probability satisfies p(c_m,c_n) ≥ (1-1/e) min{1, (M²/C²)p(m,n)}, which makes the cluster graph much denser than the original graph; this is what lets the consensus argument run under edge probabilities bounded away from 1. Information flows through a three-layer estimator: local means computed at the burn-in, cluster means obtained by averaging within a cluster, and a global estimator formed by a weighted combination of cluster means from other clusters together with the local cluster information. The l-periodic connectivity condition, defined via composition of l consecutive graphs, guarantees that within l time steps every cluster hears from every other, and the consensus lemmas turn this into a bound on the realized sample counts.

What would settle it

Run UCB-SBM in a two-cluster system with M=40 agents, K=2 arms, sub-Gaussian rewards with known gaps Δ, and choose the inter-cluster edge probability q below the Theorem 8 lower bound while keeping the within-cluster probability high enough to make each cluster internally connected. If the measured per-agent regret still behaves as (C/M)·(4C₁ log T/Δ²), the connectivity assumption is sufficient but not necessary; if the regret jumps to order M log T, the assumption is load-bearing.

Watch

Extended reading notes

Core claim

The paper claims that the system-wide per-agent regret of a cooperative multi-armed bandit on a stochastic block model can be made to grow as O((C/M) log T) rather than scaling with the number of agents M. Concretely, Theorem 8 states that with probability at least 1-7ε, the conditional expected regret is at most L plus a sum over suboptimal arms of Δ_i times the maximum of (C/M)(4C₁ log T/Δ_i²) and 2(K²+MK), plus additive constants; the coefficient C/M is exactly the degree of heterogeneity h = C/M. The authors support this with an algorithm that builds three layers of estimators — local, cluster, and global — and that only requires the cluster-level graph and the within-cluster subgraph to be l-periodically connected with high probability, a condition met when inter- and intra-cluster edge probabilities exceed factorial lower bounds that stay strictly below 1 as T grows.

Load-bearing premise

For the bound to hold, the random graph must stay l-periodically connected with high probability at every time, meaning the edge probabilities both within and across clusters must exceed the factorial lower bounds stated in Theorem 8 and every cluster must contain enough agents for those expressions to be defined; if the graph is too sparse, information cannot propagate and the O((C/M) log T) regret no longer follows.

Editorial extensions

If this is right

  • Per-agent regret coefficient becomes C/M, so in large systems with few clusters the log T term is essentially independent of the number of agents; total regret grows with C, not M.
  • Setting C=1 recovers the homogeneous case with O(log T / M) per-agent regret, and C=M recovers the fully heterogeneous setting, so the bound continuously interpolates between the two regimes.
  • The edge-probability lower bounds are strictly below 1 even as T→∞, unlike prior work whose requirement approaches 1; in particular the inter-cluster bound involves e/(e-1) times (C²/M²) times a factorial decay factor.
  • With unknown clusters, the same O((C/M) log T) regret holds with probability 1 - 7ε - 1/poly(M), provided the burn-in is extended by O(M) steps and cluster detection succeeds under a signal-to-noise condition.
  • Choosing the periodic-connectivity window l optimally between 2 and C-1 gives the weakest edge-probability assumption; Theorem 8 extends this to intra-cluster edges, so even within clusters the graph need not be complete.

Reading between the lines

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

  • The C/M coefficient suggests a direct statistical economy of scale: if cluster sizes are balanced, doubling the number of agents within each cluster while keeping C fixed halves the per-agent regret, so homogeneity is a resource that substitutes for exploration.
  • Because the analysis only needs l-periodic connectivity rather than instantaneous connectivity, the proof plausibly carries over to other random graph processes with similar mixing properties, such as temporally correlated stochastic block models.
  • For unbalanced clusters the paper suggests using the minimum cluster size; an alternative weighting by actual cluster sizes would yield a regret bound depending on the harmonic mean of cluster sizes, which the paper does not analyze.
  • The experimental finding that regret first rises then falls with p and q suggests an intermediate regime where confidence terms and connectivity trade off; formalizing the dependence of C₁ on p and q is a natural next step.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 6 minor

Summary. The paper introduces a heterogeneous multi-agent multi-armed bandit model on stochastic block models, where cluster structure determines both graph edge probabilities and reward means. It proposes UCB-SBM, an algorithm with a burn-in period and cluster-level information aggregation, and claims instance-dependent regret upper bounds of order O((C/M) log T) for known and unknown cluster settings, with edge-probability assumptions bounded away from 1. The main theoretical results are Theorem 7 and Theorem 8 (and the unknown-cluster extension Corollary 11), supported by a consensus/connectivity analysis of cluster-level subgraphs. Numerical experiments compare the algorithm against several baselines.

Significance. If the theoretical claims are correct, the paper makes a useful conceptual advance: it unifies homogeneous and fully heterogeneous cooperative bandits, exposes the degree of heterogeneity through the C/M factor, and relaxes the near-1 edge-probability assumptions of prior work. The algorithmic idea of aggregating at the cluster level is natural and the experimental section suggests clear practical gains. The paper is not self-contained, however, and the central relaxed-assumption theorem currently has a load-bearing proof error, so the main claims are not yet established as written.

major comments (1)
  1. [Sections 6.2 and Appendix E.2–E.6] Several load-bearing steps are asserted rather than proved. Theorem 2 says the proof parallels Theorem 2 of [58] and lists Propositions 1–6 as following from [58]; Lemma 16 cites Lemma 3 of [64]; Lemmas 18 and 20 cite Lemma 10 of [64] without stating the referenced result or its assumptions. Because the graph model here is an SBM rather than an Erdős–Rényi graph, the reduction to [58] needs an explicit verification that the connectivity, transmission-gap, unbiasedness, and concentration propositions remain valid under agent-dependent edge probabilities. Please include the adapted proofs, or state precisely which statements in [58]/[64] are reused and why they apply to the present setting.
minor comments (6)
  1. [Appendix A.1] The heading "Burn-in Peirod" contains a typo; it should read "Burn-in Period."
  2. [Throughout] The name "Erdos-Renyi" should be typeset as "Erdős–Rényi."
  3. [Section 1.1] The phrase "the lower bound on the edge probability in our case can be at most ..." is confusing: the requirement is an upper bound on the minimal allowed edge probability, not a lower bound on p. Please rephrase.
  4. [Table 1] Table 1 contains broken entries, including "N /A" in the C=1 row and unclear superscript/footnote symbols; the table should be regenerated for readability.
  5. [Theorem 8] The notation |c_M| is used interchangeably with c_M and |c_m|; please define explicitly that c_M = M/C and |c_m| is the size of cluster m, and use the notations consistently.
  6. [Section 8] In the discussion of Fig. 1d, the text states that UCB-SBM and DrFed-UCB "achieve the same regret" when C=M=10; since the paper's headline claim is an improvement, please clarify whether this is an expected consistency or a limitation of the algorithm.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the O((C/M) log T) regret bound is derived from model assumptions and cluster-aggregation algebra, not fitted or self-referential.

full rationale

The derivation chain is not circular. The central regret coefficient C/M is not an input to the model or algorithm; it is an algebraic consequence of balanced cluster aggregation. In the proof of Theorem 3, Proposition 12 bounds the cluster-level count E[N_{m,k}(T)|A], and the passage to per-agent pulls uses the identity N_{m,k} = |c_M| n_{m,k} = (M/C) n_{m,k}, yielding E[n_{m,k}] <= (C/M) E[N_{m,k}]. This is a derived inequality, not a fitted parameter renamed as a prediction. The graph-side assumptions are proved in the paper from the SBM edge probabilities via Lemma 4, Lemma 6, and Proposition 20, and the l-periodic connectivity analysis is original to this paper. The paper does rely frequently on prior work, e.g., 'The proof of Proposition 1 - 6 is presented in Appendix in [58] with Erdos-Renyi Models,' and [58] is first-author's prior work, but the imported propositions concern the fully heterogeneous Erdos-Renyi baseline and have stated assumptions that do not include the cluster-structure result; thus the citation is external support rather than a self-referential load-bearing premise. No uniqueness theorem from the authors is invoked to forbid alternatives, and no ansatz is smuggled in via citation. The possible factorial-power exponent issue in Proposition 20, if real, is a proof-correctness gap, not a circular dependence of the conclusion on its own assumptions. Overall, no equation is defined in terms of the claimed bound, and no fitted constant forces the log T coefficient.

Assumptions & free parameters 3 free parameters · 7 assumptions · 0 invented entities

The central claim rests on standard stochastic bandit assumptions plus the new SBM cluster model. The most significant upstream dependencies are the correctness of [58]'s ER-graph proofs and [64]'s periodic-connectivity lemma, and the strong connectivity and balance assumptions needed for cluster-level consensus to hold.

free parameters (3)
  • C1 = max{4(M+2)(1-(1-c0)/(2(M+2)))^2/(3M(1-c0)), (M+2)(1+4M d^2)/M}
    Exploration constant in the UCB index and the moment generating function bound; chosen by hand to make the concentration proof close.
  • L = (C/M) max{ln(T/(2ε))/(2δ^2), 4K log^2 T / c0}
    Burn-in period length; set to ensure initial reward estimators have enough samples before the learning phase.
  • c0 = c0(K, min_i Δ_i, M, ε, δ), existence asserted
    Constant in the transmission-gap propositions, asserted to exist but not given explicitly.
assumptions (7)
  • domain assumption Rewards are σ²-sub-Gaussian with time-invariant means.
    Section 3 states that the reward of arm k at agent i at time t follows a σ²-sub-Gaussian distribution.
  • domain assumption Clusters are balanced, or the analysis uses the smallest cluster size.
    Section 6.2 assumes a balanced model and says imbalanced cases can be handled using the minimum cluster size, which can make the cluster size 1 and invalidate some formulas.
  • domain assumption For Theorems 3, 5, and 7, within-cluster edge probability p(m,m)=1; Theorem 8 relaxes this to a lower bound.
    Stated directly in the theorem statements and in Table 1.
  • domain assumption The cluster-level subgraph is l-periodically connected with high probability, derived from the lower bounds on p(m,n).
    This is the channel for consensus; it is used in Lemma 16/18 and Proposition 20 to guarantee information propagation.
  • standard math The proof machinery of [58] and [64] is correct and applies when the minimum edge probability satisfies their conditions.
    The paper explicitly reduces several propositions to results in [58] and [64] without re-deriving them, so the correctness of those external proofs is assumed.
  • domain assumption For unknown clusters, rewards are Gaussian with known variance σ², and the CSSBM exact-recovery conditions SNR > 2 log M and C^3 ≤ SNR·δ hold.
    Section 7.2, Lemma 10, imports the exact-recovery guarantee of [12] under these signal-to-noise conditions.
  • domain assumption The graph is sampled i.i.d. from the stochastic block model at each time step.
    Definition 1 in Section 3 defines the graph process as i.i.d. samples from the SBM.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Heterogeneous Multi-agent Multi-armed Bandits on Stochastic Block Models." pith.science (2026). https://pith.science/paper/WTS5TU7C

@misc{pith2026250208003,
  author       = {Pith},
  title        = {Pith review of: Heterogeneous Multi-agent Multi-armed Bandits on Stochastic Block Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WTS5TU7C}},
  note         = {Machine review of arXiv:2502.08003}
}
abstract

We study a novel heterogeneous multi-agent multi-armed bandit problem with a cluster structure induced by stochastic block models, influencing not only graph topology, but also reward heterogeneity. Specifically, agents are distributed on random graphs based on stochastic block models - a generalized Erdos-Renyi model with heterogeneous edge probabilities: agents are grouped into clusters (known or unknown); edge probabilities for agents within the same cluster differ from those across clusters. In addition, the cluster structure in stochastic block model also determines our heterogeneous rewards. Rewards distributions of the same arm vary across agents in different clusters but remain consistent within a cluster, unifying homogeneous and heterogeneous settings and varying degree of heterogeneity, and rewards are independent samples from these distributions. The objective is to minimize system-wide regret across all agents. To address this, we propose a novel algorithm applicable to both known and unknown cluster settings. The algorithm combines an averaging-based consensus approach with a newly introduced information aggregation and weighting technique, resulting in a UCB-type strategy. It accounts for graph randomness, leverages both intra-cluster (homogeneous) and inter-cluster (heterogeneous) information from rewards and graphs, and incorporates cluster detection for unknown cluster settings. We derive optimal instance-dependent regret upper bounds of order $\log{T}$ under sub-Gaussian rewards. Importantly, our regret bounds capture the degree of heterogeneity in the system (an additional layer of complexity), exhibit smaller constants, scale better for large systems, and impose significantly relaxed assumptions on edge probabilities. In contrast, prior works have not accounted for this refined problem complexity, rely on more stringent assumptions, and exhibit limited scalability.

Figures

Figures reproduced from arXiv: 2502.08003 by the authors.

Figure 1
Figure 1. The regret of different methods across different settings [PITH_FULL_IMAGE:figures/full_fig_p020_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 63 canonical work pages

  1. [58]

    Xu and D

    M. Xu and D. Klabjan. Decentralized randomly distributed multi-agent multi-armed bandit with heterogeneous rewards. Advances on Neural Information Processing Systems , 2023

    work page 2023

  2. [64]

    Zhu and J

    J. Zhu and J. Liu. Distributed multi-armed bandits. IEEE Transactions on Automatic Control , 2023

    work page 2023

  3. [1]

    E. Abbe. Community detection and stochastic block models: recent developments. Journal of Machine Learning Research, 18(177):1–86, 2018

    work page 2018

  4. [2]

    E. Abbe, A. S. Bandeira, and G. Hall. Exact recovery in the stochastic block model. IEEE Transactions on information theory, 62(1):471–487, 2015

    work page 2015

  5. [3]

    E. Abbe, J. Fan, and K. Wang. An lp theory of pca and spectral clustering. The Annals of Statistics, 50(4):2359–2385, 2022

    work page 2022

  6. [4]

    Agarwal, V

    M. Agarwal, V. Aggarwal, and K. Azizzadenesheli. Multi-agent multi-armed bandits with limited communication. The Journal of Machine Learning Research , 23(1):9529–9552, 2022

    work page 2022

  7. [5]

    E. M. Airoldi, D. M. Blei, S. E. Fienberg, E. P. Xing, and T. Jaakkola. Mixed membership stochastic block models for relational data with application to protein-protein interactions. In Proceedings of the international biometrics society annual meeting, volume 15, page 1, 2006

    work page 2006

  8. [6]

    P. Auer, N. Cesa-Bianchi, and P. Fischer. Finite-time analysis of the multiarmed bandit problem. Machine Learning, 47 (2-3):235–256, 2002

    work page 2002

Show all 66 references
  1. [7]

    P. Auer, N. Cesa-Bianchi, Y. Freund, and R. E. Schapire. The nonstochastic multiarmed bandit problem. SIAM Journal on Computing, 32(1):48–77, 2002

    2002

  2. [8]

    Y. Ban, Y. Qi, T. Wei, L. Liu, and J. He. Meta clustering of neural bandits. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages 95–106, 2024

    2024

  3. [9]

    Battiston and M

    S. Battiston and M. Catanzaro. Statistical properties of corporate board and director networks. The European Physical Journal B, 38:345–352, 2004

    2004

  4. [10]

    Bistritz and A

    I. Bistritz and A. Leshem. Distributed multi-player bandits-a game of thrones approach.Advances in Neural Information Processing Systems, 31, 2018

    2018

  5. [11]

    Blaser, C

    E. Blaser, C. Li, and H. Wang. Federated linear contextual bandits with heterogeneous clients. In International Conference on Artificial Intelligence and Statistics , pages 631–639. PMLR, 2024

    2024

  6. [12]

    Braun, H

    G. Braun, H. Tyagi, and C. Biernacki. An iterative clustering algorithm for the contextual stochastic block model with optimality guarantees. In International Conference on Machine Learning , pages 2257–2291. PMLR, 2022

    2022

  7. [13]

    Chawla, A

    R. Chawla, A. Sankararaman, A. Ganesh, and S. Shakkottai. The gossiping insert-eliminate algorithm for multi-agent bandits. In International conference on artificial intelligence and statistics , pages 3471–3481. PMLR, 2020

    2020

  8. [14]

    L. Chen, J. Xu, S. Ren, and P. Zhou. Spatio–temporal edge service placement: A bandit learning approach. IEEE Transactions on Wireless Communications, 17(12):8388–8401, 2018

    2018

  9. [15]

    Cugmas, F

    M. Cugmas, F. Mali, and A. Žiberna. Scientific collaboration of researchers and organizations: a two-level blockmodeling approach. Scientometrics, 125(3):2471–2489, 2020

    2020

  10. [16]

    X. Dai, Z. Zhang, P. Yang, Y. Xu, X. Liu, and J. C. Lui. Axiomvision: Accuracy-guaranteed adaptive visual model selection for perspective-aware video analytics. In Proceedings of the 32nd ACM International Conference on Multimedia , pages 7229–7238, 2024

    2024

  11. [17]

    F. Delarue. Mean field games: A toy model on an Erdös-Renyi graph. ESAIM: Proceedings and Surveys , 60:1–26, 2017

    2017

  12. [18]

    Deshpande, S

    Y. Deshpande, S. Sen, A. Montanari, and E. Mossel. Contextual stochastic block models.Advances in Neural Information Processing Systems, 31, 2018

    2018

  13. [19]

    Dreveton, F

    M. Dreveton, F. Fernandes, and D. Figueiredo. Exact recovery and bregman hard clustering of node-attributed stochastic block model. Advances in Neural Information Processing Systems , 36, 2024

    2024

  14. [20]

    Dubey and A

    A. Dubey and A. Pentland. Cooperative multi-agent bandits with heavy tails. In International Conference on Machine Learning, 2730–2739, 2020

    2020

  15. [21]

    Duchemin

    Q. Duchemin. Reliable prediction in the markov stochastic block model. ESAIM: Probability and Statistics , 27:80–135, 2023

    2023

  16. [22]

    A. El Haj. Community detection in multiplex continous weighted nodes networks using an extension of the stochastic block model. Computing, 106(11):3711–3725, 2024

    2024

  17. [23]

    ERDdS and A

    P. ERDdS and A. R&wi. On random graphs i. Publ. math. debrecen, 6(290-297):18, 1959

    1959

  18. [24]

    Gentile, S

    C. Gentile, S. Li, and G. Zappella. Online clustering of bandits. In International conference on machine learning , pages 757–765. PMLR, 2014

    2014

  19. [25]

    Gentile, S

    C. Gentile, S. Li, P. Kar, A. Karatzoglou, G. Zappella, and E. Etrue. On context-dependent clustering of bandits. In International Conference on machine learning , pages 1253–1262. PMLR, 2017

    2017

  20. [26]

    P. W. Holland, K. B. Laskey, and S. Leinhardt. Stochastic blockmodels: First steps. Social networks, 5(2):109–137, 1983

    1983

  21. [27]

    Huang, W

    R. Huang, W. Wu, J. Yang, and C. Shen. Federated linear contextual bandits. Advances in Neural Information Processing Systems, 34:27057–27068, 2021

    2021

  22. [28]

    Jiang and H

    F. Jiang and H. Cheng. Multi-agent bandit with agent-dependent expected rewards. Swarm Intelligence, 1–33, 2023

    2023

  23. [29]

    Korda, B

    N. Korda, B. Szorenyi, and S. Li. Distributed clustering of linear bandits in peer to peer networks. In International conference on machine learning , pages 1301–1309. PMLR, 2016. Heterogeneous Multi-agent Multi-armed Bandit on Stochastic Block Models 23

    2016

  24. [30]

    Landgren, V

    P. Landgren, V. Srivastava, and N. E. Leonard. On distributed cooperative decision-making in multiarmed bandits. In 2016 European Control Conference . 243–248. IEEE, 2016

    2016

  25. [31]

    Landgren, V

    P. Landgren, V. Srivastava, and N. E. Leonard. Distributed cooperative decision-making in multiarmed bandits: Frequentist and Bayesian algorithms. In 2016 IEEE 55th Conference on Decision and Control . 167–172. IEEE, 2016

    2016

  26. [32]

    Landgren, V

    P. Landgren, V. Srivastava, and N. E. Leonard. Distributed cooperative decision making in multi-agent multi-armed bandits. Automatica, 125:109445, 2021

    2021

  27. [33]

    Q. Li, C. Zhao, T. Yu, J. Wu, and S. Li. Clustering of conversational bandits with posterior sampling for user preference learning and elicitation. User Modeling and User-Adapted Interaction , 33(5):1065–1112, 2023

    2023

  28. [34]

    Li and S

    S. Li and S. Zhang. Online clustering of contextual cascading bandits. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 32, 2018

    2018

  29. [35]

    S. Li, C. Gentile, A. Karatzoglou, and G. Zappella. Online context-dependent clustering in recommendations based on exploration-exploitation algorithms. ArXiv, abs/1608.03544, 2016

    2016 arXiv

  30. [36]

    S. Li, A. Karatzoglou, and C. Gentile. Collaborative filtering bandits. In Proceedings of the 39th International ACM SIGIR conference on Research and Development in Information Retrieval , pages 539–548, 2016

    2016

  31. [37]

    S. Li, W. Chen, and K.-S. Leung. Improved algorithm on online clustering of bandits. arXiv preprint arXiv:1902.09162, 2019

    1902 arXiv

  32. [38]

    Li and L

    T. Li and L. Song. Privacy-preserving communication-efficient federated multi-armed bandits. IEEE Journal on Selected Areas in Communications, 40(3):773–787, 2022

    2022

  33. [39]

    Z. Li, M. Liu, X. Dai, and J. Lui. Demystifying online clustering of bandits: Enhanced exploration under stochastic and smoothed adversarial contexts. arXiv preprint arXiv:2501.00891, 2025

    2025 arXiv

  34. [40]

    F. W. Lima, A. O. Sousa, and M. Sumuor. Majority-vote on directed Erdős–Rényi random graphs. Physica A: Statistical Mechanics and its Applications, 387(14):3503–3510, 2008

    2008

  35. [41]

    X. Liu, H. Zhao, T. Yu, S. Li, and J. C. Lui. Federated online clustering of bandits. InUncertainty in Artificial Intelligence, pages 1221–1231. PMLR, 2022

    2022

  36. [42]

    Martínez-Rubio, V

    D. Martínez-Rubio, V. Kanade, and P. Rebeschini. Decentralized cooperative stochastic bandits. Advances in Neural Information Processing Systems, 32, 2019

    2019

  37. [43]

    Mitra, H

    A. Mitra, H. Hassani, and G. Pappas. Exploiting heterogeneity in robust federated best-arm identification. arXiv preprint arXiv:2109.05700, 2021

    2021 arXiv

  38. [44]

    T. T. Nguyen and H. W. Lauw. Dynamic clustering of contextual multi-armed bandits. In Proceedings of the 23rd ACM international conference on conference on information and knowledge management , pages 1959–1962, 2014

    1959

  39. [45]

    S. Pal, A. Suggala, K. Shanmugam, and P. Jain. Blocked collaborative bandits: online collaborative filtering with per-item budget constraints. Advances in Neural Information Processing Systems , 36, 2024

    2024

  40. [46]

    C. Réda, S. Vakili, and E. Kaufmann. Near-optimal collaborative learning in bandits. In 2022-36th Conference on Neural Information Processing System, 2022

    2022

  41. [47]

    Roman, J

    R. Roman, J. Zhou, and J. Lopez. On the features and challenges of security and privacy in distributed internet of things. Computer networks, 57(10):2266–2279, 2013

    2013

  42. [48]

    Sankararaman, A

    A. Sankararaman, A. Ganesh, and S. Shakkottai. Social learning in multi agent multi armed bandits. Proceedings of the ACM on Measurement and Analysis of Computing Systems , 3(3):1–35, 2019

    2019

  43. [49]

    Stanley, T

    N. Stanley, T. Bonacci, R. Kwitt, M. Niethammer, and P. J. Mucha. Stochastic block models with multiple continuous attributes. Applied Network Science, 4:1–22, 2019

    2019

  44. [51]

    P.-A. Wang, A. Proutiere, K. Ariu, Y. Jedra, and A. Russo. Optimal algorithms for multiplayer multi-armed bandits. In International Conference on Artificial Intelligence and Statistics , pages 4120–4129. PMLR, 2020

    2020

  45. [52]

    Q. Wang, C. Zeng, W. Zhou, T. Li, S. S. Iyengar, L. Shwartz, and G. Y. Grabarnik. Online interactive collaborative filtering using multi-armed bandit with dependent arms. IEEE Transactions on Knowledge and Data Engineering , 31(8): 1569–1580, 2019. doi: 10.1109/TKDE.2018.2866041

    2019

  46. [53]

    X. Wang, L. Yang, Y.-Z. J. Chen, X. Liu, M. Hajiesmaili, D. Towsley, and J. C. Lui. Achieving near-optimal individual regret & low communications in multi-agent bandits. In The Eleventh International Conference on Learning Representations , 2022

    2022

  47. [54]

    X. Wang, L. Yang, Y.-Z. J. Chen, X. Liu, M. Hajiesmaili, D. Towsley, and J. C. Lui. Achieve near-optimal individual regret &amp; low communications in multi-agent bandits. In International Conference on Learning Representations , 2023

    2023

  48. [55]

    Z. Wang, C. Zhang, M. K. Singh, L. Riek, and K. Chaudhuri. Multitask bandit learning through heterogeneous feedback aggregation. In International Conference on Artificial Intelligence and Statistics , 1531–1539, 2021

    2021

  49. [56]

    J. Wu, C. Zhao, T. Yu, J. Li, and S. Li. Clustering of conversational bandits for user preference learning and elicitation. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management , pages 2129–2139, 24 Mengfan Xu, Liren Shan, Fatemeh Ghaff...

    2021

  50. [57]

    Xu and D

    M. Xu and D. Klabjan. Regret lower bounds in multi-agent multi-armed bandit. arXiv preprint arXiv:2308.08046, 2023

    2023 arXiv

  51. [59]

    Z. Yan, Q. Xiao, T. Chen, and A. Tajer. Federated multi-armed bandit via uncoordinated exploration. InIEEE International Conference on Acoustics, Speech and Signal Processing . 5248–5252. IEEE, 2022

    2022

  52. [60]

    H. Yang, X. Liu, Z. Wang, H. Xie, J. C. Lui, D. Lian, and E. Chen. Federated contextual cascading bandits with asynchronous communication and heterogeneous users. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 38, pages 20596–20603, 2024

    2024

  53. [61]

    P. Yang, N. Zhang, S. Zhang, L. Yu, J. Zhang, and X. Shen. Content popularity prediction towards location-aware mobile edge caching. IEEE Transactions on Multimedia , 21(4):915–929, 2018

    2018

  54. [62]

    W. W. Zachary. An information flow model for conflict and fission in small groups. Journal of anthropological research, 33(4):452–473, 1977

    1977

  55. [63]

    Y. Zhao, J. Zhao, L. Jiang, R. Tan, D. Niyato, Z. Li, L. Lyu, and Y. Liu. Privacy-preserving blockchain-based federated learning for iot devices. IEEE Internet of Things Journal , 8(3):1817–1829, 2020

    2020

  56. [65]

    J. Zhu, R. Sandhu, and J. Liu. A distributed algorithm for sequential decision making in multi-armed bandit with homogeneous rewards. In 59th IEEE Conference on Decision and Control . 3078–3083. IEEE, 2020

    2020

  57. [66]

    J. Zhu, E. Mulle, C. S. Smith, and J. Liu. Decentralized multi-armed bandit can outperform classic upper confidence bound. arXiv preprint arXiv:2111.10933, 2021

    2021 arXiv

  58. [67]

    Z. Zhu, J. Zhu, J. Liu, and Y. Liu. Federated bandit: A gossiping approach. In Abstract Proceedings of the 2021 ACM SIGMETRICS/International Conference on Measurement and Modeling of Computer Systems , 3–4, 2021. Heterogeneous Multi-agent Multi-armed Bandit on Stochastic Block...

    2021

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.