Intelligent Channel Allocation for IEEE 802.11be Multi-Link Operation: When MAB Meets LLM

Jingwen Tong; Jun Zhang; Liqun Fu; Shumin Lian

REVIEW 4 major objections 5 minor 37 references

Intelligent Channel Allocation for IEEE 802.11be Multi-Link Operation: When MAB Meets LLM

T0 review · 4 major / 5 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 A best-arm Monte Carlo tree search learns near-optimal WiFi 7 channel assignments online, and an LLM warm start makes it converge over 63% faster in dense networks.

desk verdict The algorithm and experiments are worth a look, but the main theorem's Gaussian-and-bounded assumption is a flat contradiction; major revision needed. read the letter →

arxiv 2506.04594 v1 pith:ZFM6ACZW submitted 2025-06-05 cs.NI cs.AIeess.SP

Shumin Lian , Jingwen Tong , Jun Zhang , Liqun Fu This is my paper · ORCID

classification cs.NIcs.AIeess.SP

keywords WiFi7IEEE802.11bemulti-linkoperationchannelallocationmulti-armedbanditsbest-armidentificationMonteCarlotreesearchlargelanguagemodels

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 tries to establish that the hardest part of IEEE 802.11be Multi-Link Operation—deciding which channels each STA should use when many APs contend in a dense network—can be solved online, without knowing the network in advance, by treating each complete channel assignment as an arm of a multi-armed bandit. It proposes BAI-MCTS, a Monte Carlo tree search that applies a best-arm identification rule at every tree layer, and proves that it returns an $\epsilon$-optimal allocation with probability at least $1-\delta$ while spending at most a sum of per-layer characteristic times on sampling. To cope with the exponential arm space, the paper adds LLM-BAI-MCTS, which uses a large language model prompted with few examples to fix some STAs' channels in advance and shrink the tree. The numerical claim is that BAI-MCTS converges about 50.44% faster than the DNG-MCTS baseline at 98% of optimal throughput, and LLM-BAI-MCTS converges over 63.32% faster than BAI-MCTS in dense networks; if true, this makes guaranteed near-optimal MLO channel allocation practical in settings where exhaustive search over $7^6=117{,}649$ configurations is infeasible.

What carries the argument

The load-bearing object is the layer-wise EB-TC$\epsilon$ rule—the Empirical-Best and Transportation-Cost $\epsilon$ best-arm identification algorithm, here applied independently at each node of the MCTS tree. It is what converts the tree search from a heuristic into a fixed-confidence PAC procedure: at each node it identifies the two most promising children, allocates samples between them, and declares the layer converged only when the gap between the empirical best and every challenger, inflated by $\epsilon/N$, clears a generalized-likelihood-ratio threshold. The LLM component plays a second, support role: in-context learning plus chain-of-thought prompting produces an initial assignment for a subset of STAs, removing those layers from the tree and reducing the sample-complexity bound correspondingly.

What would settle it

Run BAI-MCTS on the paper's 3-AP/6-STA network with a fixed $\epsilon=0.02$, several decreasing values of $\delta$, and enough repetitions to estimate $E[\tau_{\epsilon,\delta}]$; if $E[\tau_{\epsilon,\delta}]/\log(1/\delta)$ fails to approach the sum of the layer-wise characteristic times from Theorem 1, or if the fraction of runs that end below $98\%$ of the exhaustive-search optimum exceeds $\delta$, the central guarantee is refuted.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a hierarchical best-arm identification procedure can replace the heuristic exploration of MCTS with a PAC guarantee without losing the tree's ability to navigate a huge arm space. At each node of the Monte Carlo tree, the algorithm maintains the empirical best child and a 'transportation-cost challenger,' samples between them adaptively, and applies a generalized-likelihood-ratio stopping rule; once a layer stops, the search is constrained to the winner, and the path of winners to the root is the recommended channel allocation. The paper proves (Theorem 1) that, for Gaussian rewards bounded in $[0,1]$, the asymptotic expected sample complexity divided by $\log(1/\delta)$ is bounded by the sum over layers of the maximum characteristic time of that layer, and (Theorem 2) that the probability of recommending a non-$\epsilon$-optimal arm before stopping decays exponentially in the number of selections of the parent nodes. The LLM variant is not given a separate guarantee; it is presented as a way to cut the tree height and therefore the same bound, with the empirical claim that few-shot prompted LLMs produce initial allocations good enough to speed convergence by over 63% in dense networks.

Load-bearing premise

The load-bearing premise is that at each node of the search tree, the rewards of its children are independent draws from fixed distributions once the layers below have converged; before that convergence, those rewards are not independent because they depend on the random choices made in lower layers.

Editorial extensions

If this is right

If the guarantees hold, a central coordinator can deploy BAI-MCTS with no prior knowledge of the carrier-sensing graph and still end up within $\epsilon$ of the best possible throughput with probability at least $1-\delta$.
The sample-complexity bound scales as the sum of per-layer characteristic times rather than with the total number of arms, so the algorithm remains applicable when the arm count is exponential in the number of AP-STA pairs.
Using an LLM to pre-assign $L$ STAs is equivalent to reducing the tree height by $L$ layers; the paper's numbers indicate this preserves or improves throughput while cutting convergence time, with $L=16$ of 24 STAs outperforming both pure BAI-MCTS and a full-LLM assignment.
In the tested settings, STR-mode MLO with BAI-MCTS allocation beats SLO and channel bonding in throughput and cuts average latency by about 36.83% relative to SLO, giving a practical reason to prefer MLO when the coordinator can learn the allocation.

Reading between the lines

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

Because the Theorem 1 proof rests on a layer-wise i.i.d. assumption that becomes valid only after lower layers have converged, a practical deployment should monitor convergence indicators before trusting the bound; one testable extension is to measure the serial correlation of child-node rewards before parent convergence and check whether it inflates the true sample complexity.
The LLM warm start is best read as a transfer-learning accelerator rather than a new guarantee; the same template could be tested on other tree-structured wireless allocation problems, such as link activation or AP-STA pairing, where a cheap initializer would shrink the search tree.
The reported differences between three commercial LLMs are small, which suggests a testable extension: whether a much smaller open-weight model, or even a rule-based heuristic distilled from the CoT steps, recovers most of the 63% speedup without external API calls.
The observed performance drop when $L=24$ (all STAs assigned by the LLM) compared with $L=16$ hints that the optimal amount of LLM intervention is an interior value; a sensitivity analysis over $L$ with confidence intervals would turn this into a design guideline.

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.

Desk Editor's Note

read the letter

Dear colleague,

The short version: this is a sensible engineering paper with a promising algorithm, but the main theoretical guarantee is unsupported because Theorem 1 assumes rewards are both bounded in [0,1] and Gaussian, which can't both be true. I'd send it back for major revision rather than desk-reject it, but the theorem needs to be fixed.

What's actually new: the authors combine best-arm identification with Monte Carlo tree search for IEEE 802.11be multi-link channel allocation, and then use an LLM to pre-assign some stations and shrink the search tree. That's a reasonable incremental combination, and they include working code. The experiments show BAI-MCTS converging faster than UCT and DNG-MCTS on a small 3-AP/6-STA benchmark, and the LLM variant shaves off more time when it fixes L stations. Those results are plausible.

The soft spots are real. The Gaussian-vs-bounded contradiction appears in Theorem 1 (Section V-C) and again in Appendix A, where they say rewards are 'independent Gaussian distributions with values bounded within the interval [0,1]' and then apply Theorem 1 of [36]. If you clip or truncate the Gaussian, the characteristic times in Eqs. (24)-(25) no longer hold as derived. The paper gives no alternative proof. That's the load-bearing issue. There's also a secondary gap: the layer-by-layer recursion leans on Lemma 23 of [36], which controls allocation proportions, not the reward distribution induced by lower-layer randomization, so the i.i.d. assumption at each node isn't established.

The numerics are okay but not airtight: no error bars despite 10^3 MC experiments, and the LLM comparison is partly by construction because fixing L stations reduces tree height. The headline percentages are point estimates from one configuration. I also think the comparison to DNG-MCTS is a bit weak, but UCT is a reasonable baseline.

Recommendation: worth a serious referee, but not as is. The authors need to either repair the theorem (e.g., state a truncated Gaussian and re-derive or cite the right bound) or drop the sample-complexity claim. The algorithmic story and code are worth keeping.

For me: I might bring it to reading group as a cautionary tale, but I wouldn't cite it this year. It's not incoherent in its engineering parts, but the theory is internally contradictory on its own terms.

Referee Report

4 major / 5 minor

Summary. The paper studies dynamic channel allocation in dense IEEE 802.11be networks with Multi-Link Operation (MLO) under the STR mode. It formulates the allocation problem as a stochastic multi-armed bandit (MAB) with a large arm space, proposes a Best-Arm Identification-enabled Monte Carlo Tree Search (BAI-MCTS) algorithm, and adds an LLM-initialized variant (LLM-BAI-MCTS) that uses in-context learning and chain-of-thought prompting to reduce the search space. The main theoretical claims are an upper bound on expected sample complexity (Theorem 1, Eq. (20)) and an upper bound on recommendation error probability (Theorem 2, Eq. (21)). The numerical section reports convergence improvements of about 50.44% for BAI-MCTS over DNG-MCTS and about 63.32% for LLM-BAI-MCTS over BAI-MCTS in dense networks, together with throughput and latency comparisons across different network sizes and LLM backbones.

Significance. If the theoretical guarantees were sound, the paper would make a useful contribution to online resource allocation in WiFi 7, and the combination of MCTS with best-arm identification plus LLM-based initialization is a novel and potentially practical idea. The paper also provides a reproducible code link and a broad set of simulations, including comparisons across multiple LLMs, which are positive features. However, the central theorem is built on an internally inconsistent assumption set, and the empirical headline improvements are not accompanied by statistical confidence measures; these issues currently prevent the results from being relied upon as stated.

major comments (4)

[Section V-C, Theorem 1 (Eq. (20)) and Appendix A] The theorem assumes both that 'the reward bounded within the interval [0, 1]' and that the reward vector follows ν ∼ N^{|L|N}. A Gaussian distribution has unbounded support, so these assumptions are mutually contradictory. Appendix A repeats the same contradiction, stating that rewards 'follow independent Gaussian distributions with values bounded within the interval [0,1]'. If a truncated or clipped Gaussian is intended, the characteristic-time formulas in Eqs. (24)-(25), which are derived for exact Gaussian bandits in [36], no longer apply and no alternative derivation is supplied. This invalidates the stated upper bound in Eq. (20) and the proof of Theorem 1 as written.
[Appendix A (layer-wise recursion)] The proof assumes that, once lower layers have converged, the rewards of a node's children are approximately i.i.d. Gaussian, and it invokes Lemma 23 of [36] to justify this. That lemma only controls the allocation proportions of the EB-TC algorithm; it does not establish that the empirical reward distribution of a parent node becomes i.i.d. Gaussian while lower-layer random selections are still ongoing. The dependency between layers is not resolved, so the recursive application of the EB-TC analysis to every layer is not justified.
[Section VII, Figs. 9 and 13] The paper reports 'All results are obtained from 10^3 MC experiments' but presents only point estimates of average throughput and convergence rates, with no error bars, confidence intervals, or significance tests. In particular, the headline claims of 50.44% and 63.32% convergence improvements are single numbers without variance, so a reader cannot determine whether these differences are statistically meaningful or within Monte Carlo noise.
[Section VI-C and Fig. 13] The 63.32% improvement of LLM-BAI-MCTS over BAI-MCTS is partly by construction: for L = N − 8, the algorithm reduces the MCT height by L layers, so it solves a strictly smaller search problem. The comparison consequently conflates the quality of the LLM's initial allocation with the mechanical benefit of a reduced tree. A fairer benchmark would give BAI-MCTS the same reduced height but with random or heuristic pre-allocation of the L fixed STAs, or otherwise separate the two effects.

minor comments (5)

[Section VI, first paragraph] The text reads 'provide a high-quality initialization for the BASI-MCTS algorithm'; this appears to be a typo for 'BAI-MCTS'.
[Section VII and reference list] The baseline algorithm is referred to as 'DNG-MCTS' in the abstract and most of the text, but as 'DGN-MCTS' in Section VII and in the description of reference [37]. Please use one consistent name.
[Reference [4]] The publication month is given as 'Fed. 2022'; this should be 'Feb. 2022'.
[Section II, last paragraph] The sentence 'catalyzing breakthrough innovations across diverse domains Ref. [30]–[32]' is missing the intended bracket formatting; it should read 'across diverse domains [30]–[32]'.
[Theorem 2 and Appendix B] The notation '1 (ϵ <∆max)' is ambiguous; if it denotes an indicator function, it should be typeset as \mathbb{1}_{\{ϵ < Δmax\}} or defined explicitly.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the BAI-MCTS bound imports an external best-arm-identification theorem, and the LLM-BAI-MCTS speedup is a transparent tree-height reduction rather than a fitted prediction.

full rationale

The central theoretical results, Theorem 1 (Eq. (20)) and Theorem 2 (Eq. (21)), are obtained by applying the EB-TC epsilon fixed-confidence best-arm identification analysis of the external reference [36] layer by layer (Appendix A, Eqs. (23)-(27)), not by fitting parameters to the paper's own outputs or by a self-citation chain. Reference [36] is authored by Jourdan, Degenne, and Kaufmann, who do not overlap with the present authors, so the main guarantee rests on independent, published machinery. The empirical convergence claims are benchmarked against external baselines (UCT, DNG-MCTS, random selection, and exhaustive-search optimum value) in Section VII, so they are measured rather than constructed by definition. The self-citations that appear ([1], [13], [16], [34]) are contextual (conference precursor, LLM survey, simulator source, prior bandit application) and are not load-bearing for the claimed epsilon-optimality or sample-complexity bounds. The LLM-BAI-MCTS gain is transparently attributed in Section VI-C to reducing the MCT height by L layers: 'decreasing the height of MCT by L layers, as these LLM-determined allocations no longer require exploration within the tree structure.' That is an explicit design property, not a hidden equivalence between predicted and fitted quantities. The paper does contain a correctness-relevant inconsistency: Theorem 1 assumes rewards are simultaneously bounded in [0,1] and Gaussian, which cannot both hold exactly, and Appendix A repeats the phrase 'independent Gaussian distributions with values bounded within the interval [0,1]'. This is a serious proof-level flaw, but it is not circularity: the sample-complexity bound is not assumed as its own conclusion, and the argument does not reduce to its inputs by construction. The load-bearing assumption about lower-layer reward distributions via Lemma 23 of [36] is likewise a proof gap rather than a circular step. Overall, no specific reduction of a claimed result to its inputs could be exhibited, so the circularity score is 0.

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

The central claims rest on standard bandit theory, the ICN throughput model, and a few simulation-specific parameters. No new physical entities are introduced. The free parameters are reasonable but the theoretical assumptions about i.i.d. layer-wise samples are the most fragile part.

free parameters (3)

epsilon = 0.02
Suboptimality tolerance in the epsilon-optimal arm definition and in EB-TC and stopping rules. Chosen by the authors; results depend on it.
delta = 0.1
Confidence parameter in the PAC guarantee. Chosen by the authors; affects stopping thresholds and reported convergence.
L (number of LLM-assigned STAs) = 8, 16, 24 in different scenarios
The number of stations whose configuration is fixed by the LLM. This is a tunable design parameter that directly reduces the MCTS tree height and thus sample complexity.

assumptions (4)

ad hoc to paper Rewards are independent Gaussian distributions bounded within [0,1]
Used to apply the EB-TC analysis from [36]; Gaussian distributions are unbounded, so the boundedness assumption conflicts with the Gaussian assumption.
domain assumption Layer-wise samples of child node rewards become i.i.d. once lower layers converge
Invoked in Appendix A via Lemma 23 of [36] to justify the recursive application of the EB-TC bound; not fully proven in the MCTS setting.
domain assumption The Ideal CSMA Network (ICN) model of [11] accurately computes WiFi 7 MLO throughput
The paper adopts the CTRM/ICN model and uses it for the reward computations; its accuracy for MLO/STR is not independently validated, though a discrete event simulator is used for numerical results.
domain assumption The discrete event simulator from [16] faithfully represents the CSMA protocol in the simulated scenarios
Used to generate all numerical results; no verification of the simulator is provided in this paper.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Intelligent Channel Allocation for IEEE 802.11be Multi-Link Operation: When MAB Meets LLM." pith.science (2026). https://pith.science/paper/ZFM6ACZW

@misc{pith2026250604594,
  author       = {Pith},
  title        = {Pith review of: Intelligent Channel Allocation for IEEE 802.11be Multi-Link Operation: When MAB Meets LLM},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZFM6ACZW}},
  note         = {Machine review of arXiv:2506.04594}
}

abstract

WiFi networks have achieved remarkable success in enabling seamless communication and data exchange worldwide. The IEEE 802.11be standard, known as WiFi 7, introduces Multi-Link Operation (MLO), a groundbreaking feature that enables devices to establish multiple simultaneous connections across different bands and channels. While MLO promises substantial improvements in network throughput and latency reduction, it presents significant challenges in channel allocation, particularly in dense network environments. Current research has predominantly focused on performance analysis and throughput optimization within static WiFi 7 network configurations. In contrast, this paper addresses the dynamic channel allocation problem in dense WiFi 7 networks with MLO capabilities. We formulate this challenge as a combinatorial optimization problem, leveraging a novel network performance analysis mechanism. Given the inherent lack of prior network information, we model the problem within a Multi-Armed Bandit (MAB) framework to enable online learning of optimal channel allocations. Our proposed Best-Arm Identification-enabled Monte Carlo Tree Search (BAI-MCTS) algorithm includes rigorous theoretical analysis, providing upper bounds for both sample complexity and error probability. To further reduce sample complexity and enhance generalizability across diverse network scenarios, we put forth LLM-BAI-MCTS, an intelligent algorithm for the dynamic channel allocation problem by integrating the Large Language Model (LLM) into the BAI-MCTS algorithm. Numerical results demonstrate that the BAI-MCTS algorithm achieves a convergence rate approximately $50.44\%$ faster than the state-of-the-art algorithms when reaching $98\%$ of the optimal value. Notably, the convergence rate of the LLM-BAI-MCTS algorithm increases by over $63.32\%$ in dense networks.

Figures

Figures reproduced from arXiv: 2506.04594 by the authors.

**Figure 2.** An illustration of the SLO, NSTR, and STR transmission modes in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** (a) The carrier-sensing graph of five links on the same channel; (b) [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (8 more)

**Figure 4.** Figure 4: An illustration of the intuitions behind the proposed algorithms. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: An illustration of four steps of MCTS, containing selection, expansion, [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: An illustration of the BAI-MCTS algorithm’s channel allocation [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 7.** Figure 7: An example of the LLM-assisted channel allocation. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

**Figure 8.** Figure 8: A simple network with three BSSs distributed in a [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]

**Figure 10.** Figure 10: The selected percentage of each arm of each AP-STA pair in the [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]

**Figure 11.** Figure 11: Network throughput versus the number of STAs in the network [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]

**Figure 14.** Figure 14: The average network throughput of different algorithms when [PITH_FULL_IMAGE:figures/full_fig_p013_14.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 32 canonical work pages

[36]

An ε-best-arm identifi- cation algorithm for fixed-confidence and beyond,

M. Jourdan, R. Degenne, and E. Kaufmann, “An ε-best-arm identifi- cation algorithm for fixed-confidence and beyond,” Adv. Neural Inf. Process. Syst. , vol. 36, pp. 16 578–16 649, New Orleans, Louisiana, United States, Dec. 2023

work page 2023
[1]

Dynamic channel allocation via bandit learning for WiFi 7 networks with multi-link operation,

S. Lian, J. Tong, and L. Fu, “Dynamic channel allocation via bandit learning for WiFi 7 networks with multi-link operation,” in Proc. IEEE Wireless Commun. and Net. Conf. , Milan, Italy, May 2025

work page 2025
[2]

IEEE 802.11 wireless local area networks,

B. P. Crow, I. Widjaja, J. G. Kim, and P. T. Sakai, “IEEE 802.11 wireless local area networks,” IEEE Commun. Mag., vol. 35, no. 9, pp. 116–126, Sept. 1997

work page 1997
[3]

IEEE 802.11 be Wi-Fi 7: New challenges and opportunities,

C. Deng, X. Fang, X. Han, X. Wang, L. Yan, R. He, Y . Long, and Y . Guo, “IEEE 802.11 be Wi-Fi 7: New challenges and opportunities,” IEEE Commun. Surv. Tut. , vol. 22, no. 4, pp. 2136–2166, 4th Quarter 2020

work page 2020
[4]

Overview and performance evaluation of Wi-Fi 7,

C. Chen, X. Chen, D. Das, D. Akhmetov, and C. Cordeiro, “Overview and performance evaluation of Wi-Fi 7,” IEEE Commun. Stand. Mag. , vol. 6, no. 2, pp. 12–18, Fed. 2022

work page 2022
[5]

Wi-Fi multi-link operation: An experimental study of latency and throughput,

M. Carrascosa-Zamacois, G. Geraci, E. Knightly, and B. Bellalta, “Wi-Fi multi-link operation: An experimental study of latency and throughput,” IEEE/ACM Trans. Netw., vol. 32, no. 1, pp. 308–322, Jun. 2023

work page 2023
[6]

The impact of channel bonding on 802.11 n network management,

L. Deek, E. Garcia-Villegas, E. Belding, S.-J. Lee, and K. Almeroth, “The impact of channel bonding on 802.11 n network management,” in Proc. ACM Conf. Emerg. Netw. Exp. Technol., Tokyo, Japan, Dec. 2011, pp. 1–12

work page 2011
[7]

Where to aggregate?: a comparative study of aggregation strategies in multi-radio system,

Y . M. Saputra and J. H. Yun, “Where to aggregate?: a comparative study of aggregation strategies in multi-radio system,” Proc. Asia Pac. Wireless Commun. Symp., Seoul, Korea, Aug. 2013

work page 2013

Show all 37 references

[8]

Multi-link operation in IEEE 802.11 be WLANs,

´A. L ´opez-Ravent´os and B. Bellalta, “Multi-link operation in IEEE 802.11 be WLANs,” IEEE Wireless Commun. , vol. 29, no. 4, pp. 94– 100, Apr. 2022

2022
[9]

Delay analysis of IEEE 802.11 be multi-link operation under finite load,

B. Bellalta, M. Carrascosa, L. Galati-Giordano, and G. Geraci, “Delay analysis of IEEE 802.11 be multi-link operation under finite load,” IEEE Wireless Commun. Lett., vol. 12, no. 4, pp. 595–599, Apr. 2023

2023
[10]

IEEE 802.11 be network throughput optimization with multi-link operation and AP controller,

L. Zhang, H. Yin, S. Roy, L. Cao, X. Gao, and V . Sathya, “IEEE 802.11 be network throughput optimization with multi-link operation and AP controller,” IEEE Internet Things J. , vol. 11, no. 13, pp. 23 850–23 861, Apr. 2024

2024
[11]

A distributed CSMA algorithm for throughput and utility maximization in wireless networks,

L. Jiang and J. Walrand, “A distributed CSMA algorithm for throughput and utility maximization in wireless networks,” IEEE/ACM Trans. Netw., vol. 18, no. 3, pp. 960–972, Mar. 2009

2009
[12]

Regret analysis of stochastic and nonstochastic multi-armed bandit problems,

S. Bubeck, N. Cesa-Bianchi et al. , “Regret analysis of stochastic and nonstochastic multi-armed bandit problems,” Found. Trends Mach. Learn., vol. 5, no. 1, pp. 1–122, Jan. 2012

2012
[13]

Wirelessllm: Empowering large language models towards wireless intelligence,

J. Shao, J. Tong, Q. Wu, W. Guo, Z. Li, Z. Lin, and J. Zhang, “Wirelessllm: Empowering large language models towards wireless intelligence,” J. Commun. Information Netw. , vol. 9, no. 2, pp. 99–112, Jun. 2024

2024
[14]

A survey on in-context learning,

Q. Dong, L. Li, D. Dai, C. Zheng, Z. Wu, B. Chang, X. Sun, J. Xu, and Z. Sui, “A survey on in-context learning,” arXiv preprint arXiv:2301.00234, 2022

2022 arXiv
[15]

A survey of chain of thought reasoning: Advances, frontiers and future,

Z. Chu, J. Chen, Q. Chen, W. Yu, T. He, H. Wang, W. Peng, M. Liu, B. Qin, and T. Liu, “A survey of chain of thought reasoning: Advances, frontiers and future,” arXiv preprint arXiv:2309.15402 , 2023

2023 arXiv
[16]

Throughput enhancement of full-duplex CSMA networks using multiplayer bandits,

J. Tong, L. Fu, and Z. Han, “Throughput enhancement of full-duplex CSMA networks using multiplayer bandits,” IEEE Internet Things J. , vol. 8, no. 15, pp. 11 807–11 821, Aug. 2021

2021
[17]

An experimental study of latency for ieee 802.11 be multi-link operation,

M. Carrascosa, G. Geraci, E. Knightly, and B. Bellalta, “An experimental study of latency for ieee 802.11 be multi-link operation,” in Proc. IEEE Int. Conf. on Commun. (ICC) , Seoul, Korea, Aug. 2022

2022
[18]

Adaptive multi-link channel access under cross-link interference in next-generation WLANs,

J. Park, J. Han, and S. Bahk, “Adaptive multi-link channel access under cross-link interference in next-generation WLANs,” IEEE Internet Things J., vol. 11, no. 14, Jul. 2024

2024
[19]

WiFi 7 with different multi-link channel access schemes: Modeling, fairness and optimization,

J. Zhang, Q. Tan, Y . Gao, X. Sun, and W. Zhan, “WiFi 7 with different multi-link channel access schemes: Modeling, fairness and optimization,” IEEE Trans. on Commun. , vol. 72, no. 10, pp. 6225– 6236, Oct. 2024

2024
[20]

Channel selection for Wi-Fi 7 multi-link operation via optimistic-weighted VDN and parallel transfer reinforcement learning,

P. E. Iturria-Rivera, M. Chenier, B. Herscovici, B. Kantarci, and M. Erol- Kantarci, “Channel selection for Wi-Fi 7 multi-link operation via optimistic-weighted VDN and parallel transfer reinforcement learning,” in Proc. IEEE Int. Symp. on Pers., Indoor and Mobile Radio Commun...

2023
[21]

A federated reinforcement learning framework for link activation in multi-link Wi-Fi networks,

R. Ali and B. Bellalta, “A federated reinforcement learning framework for link activation in multi-link Wi-Fi networks,” in IEEE Int. Black Sea Conf. Commun. Netw., Istanbul, Turkiye, Jul. 2023

2023
[22]

Learning combinatorial optimization on graphs: A survey with appli- cations to networking,

N. Vesselinova, R. Steinert, D. F. Perez-Ramirez, and M. Boman, “Learning combinatorial optimization on graphs: A survey with appli- cations to networking,” IEEE Access, vol. 8, pp. 120 388–120 416, Jun. 2020

2020
[23]

The roadmap to 6G: AI empowered wireless networks,

K. B. Letaief, W. Chen, Y . Shi, J. Zhang, and Y .-J. A. Zhang, “The roadmap to 6G: AI empowered wireless networks,” IEEE commun. magazine, vol. 57, no. 8, pp. 84–90, Aug. 2019

2019
[24]

A tutorial on cross-layer opti- mization in wireless networks,

X. Lin, N. B. Shroff, and R. Srikant, “A tutorial on cross-layer opti- mization in wireless networks,” IEEE J. Sel. Areas Commun. , vol. 24, no. 8, pp. 1452–1463, Dec. 2006

2006
[25]

Deep learning in mobile and wireless networking: A survey,

C. Zhang, P. Patras, and H. Haddadi, “Deep learning in mobile and wireless networking: A survey,” IEEE Commun. Sur. & Tut. , vol. 21, no. 3, pp. 2224–2287, Third Quad 2019

2019
[26]

Combinatorial multivariant multi- armed bandits with applications to episodic reinforcement learning and beyond,

X. Liu, S. Wang, J. Zuo, H. Zhong, X. Wang, Z. Wang, S. Li, M. Hajiesmaili, J. Lui, and W. Chen, “Combinatorial multivariant multi- armed bandits with applications to episodic reinforcement learning and beyond,” arXiv preprint arXiv:2406.01386 , 2024

2024 arXiv
[27]

Fair probabilistic multi-armed bandit with applications to network optimization,

Z. Guo, C. Zhang, M. Li, and M. Krunz, “Fair probabilistic multi-armed bandit with applications to network optimization,” IEEE Trans. Mach. Learn. Commun. Netw. , vol. 2, pp. 994–1016, Jul. 2024

2024
[28]

Optimal rate sampling in 802.11 systems: Theory, design, and implementation,

R. Combes, J. Ok, A. Proutiere, D. Yun, and Y . Yi, “Optimal rate sampling in 802.11 systems: Theory, design, and implementation,” IEEE Trans. Mobile Comput. , vol. 18, no. 5, pp. 1145–1158, May 2018

2018
[29]

Best-arm identification in unimodal bandits,

R. Poiani, M. Jourdan, E. Kaufmann, and R. Degenne, “Best-arm identification in unimodal bandits,” arXiv preprint arXiv:2411.01898 , 2024

2024 arXiv
[30]

Introspective tips: Large language model for in-context decision making,

L. Chen, L. Wang, H. Dong, Y . Du, J. Yan, F. Yang, S. Li, P. Zhao, S. Qin, S. Rajmohan et al., “Introspective tips: Large language model for in-context decision making,” arXiv preprint arXiv:2305.11598 , 2023

2023 arXiv
[31]

Solving general natural-language-description optimization problems with large language models,

J. Zhang, W. Wang, S. Guo, L. Wang, F. Lin, C. Yang, and W. Yin, “Solving general natural-language-description optimization problems with large language models,” arXiv preprint arXiv:2407.07924 , 2024

2024 arXiv
[32]

Large language model (LLM)-enabled in-context learning for wireless network optimization: A case study of power control,

H. Zhou, C. Hu, D. Yuan, Y . Yuan, D. Wu, X. Liu, and C. Zhang, “Large language model (LLM)-enabled in-context learning for wireless network optimization: A case study of power control,” arXiv preprint arXiv:2408.00214, 2024

2024 arXiv
[33]

Finite-time analysis of the multiarmed bandit problem,

P. Auer, “Finite-time analysis of the multiarmed bandit problem,” Mach. Learn., vol. 47, no. 2/3, pp. 235–256, May 2002

2002
[34]

Model-based Thompson sampling for frequency and rate selection in underwater acoustic communica- tions,

J. Tong, L. Fu, Y . Wang, and Z. Han, “Model-based Thompson sampling for frequency and rate selection in underwater acoustic communica- tions,” IEEE Trans. Wireless Commun., vol. 22, no. 10, pp. 6946–6961, Oct. 2023

2023
[35]

Best arm identification in multi-armed bandits,

J.-Y . Audibert, S. Bubeck, and R. Munos, “Best arm identification in multi-armed bandits,” in Conf. on Learn. Theory , Haifa, Israel, Jun. 2010

2010
[37]

Bayesian mixture modeling and inference based Thompson sampling in Monte-Carlo tree search,

A. Bai, F. Wu, and X. Chen, “Bayesian mixture modeling and inference based Thompson sampling in Monte-Carlo tree search,” Adv. Neural Inf. Process. Syst., vol. 2, p. 1646–1654, Lake Tahoe, Nevada, Dec. 2013

2013

Pith tools

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

[1] [36]

An ε-best-arm identifi- cation algorithm for fixed-confidence and beyond,

M. Jourdan, R. Degenne, and E. Kaufmann, “An ε-best-arm identifi- cation algorithm for fixed-confidence and beyond,” Adv. Neural Inf. Process. Syst. , vol. 36, pp. 16 578–16 649, New Orleans, Louisiana, United States, Dec. 2023

work page 2023

[2] [1]

Dynamic channel allocation via bandit learning for WiFi 7 networks with multi-link operation,

S. Lian, J. Tong, and L. Fu, “Dynamic channel allocation via bandit learning for WiFi 7 networks with multi-link operation,” in Proc. IEEE Wireless Commun. and Net. Conf. , Milan, Italy, May 2025

work page 2025

[3] [2]

IEEE 802.11 wireless local area networks,

B. P. Crow, I. Widjaja, J. G. Kim, and P. T. Sakai, “IEEE 802.11 wireless local area networks,” IEEE Commun. Mag., vol. 35, no. 9, pp. 116–126, Sept. 1997

work page 1997

[4] [3]

IEEE 802.11 be Wi-Fi 7: New challenges and opportunities,

C. Deng, X. Fang, X. Han, X. Wang, L. Yan, R. He, Y . Long, and Y . Guo, “IEEE 802.11 be Wi-Fi 7: New challenges and opportunities,” IEEE Commun. Surv. Tut. , vol. 22, no. 4, pp. 2136–2166, 4th Quarter 2020

work page 2020

[5] [4]

Overview and performance evaluation of Wi-Fi 7,

C. Chen, X. Chen, D. Das, D. Akhmetov, and C. Cordeiro, “Overview and performance evaluation of Wi-Fi 7,” IEEE Commun. Stand. Mag. , vol. 6, no. 2, pp. 12–18, Fed. 2022

work page 2022

[6] [5]

Wi-Fi multi-link operation: An experimental study of latency and throughput,

M. Carrascosa-Zamacois, G. Geraci, E. Knightly, and B. Bellalta, “Wi-Fi multi-link operation: An experimental study of latency and throughput,” IEEE/ACM Trans. Netw., vol. 32, no. 1, pp. 308–322, Jun. 2023

work page 2023

[7] [6]

The impact of channel bonding on 802.11 n network management,

L. Deek, E. Garcia-Villegas, E. Belding, S.-J. Lee, and K. Almeroth, “The impact of channel bonding on 802.11 n network management,” in Proc. ACM Conf. Emerg. Netw. Exp. Technol., Tokyo, Japan, Dec. 2011, pp. 1–12

work page 2011

[8] [7]

Where to aggregate?: a comparative study of aggregation strategies in multi-radio system,

Y . M. Saputra and J. H. Yun, “Where to aggregate?: a comparative study of aggregation strategies in multi-radio system,” Proc. Asia Pac. Wireless Commun. Symp., Seoul, Korea, Aug. 2013

work page 2013