Pith. sign in

REVIEW 3 major objections 5 minor 34 references

Truthful and Trustworthy IoT AI Agents via Immediate-Penalty Enforcement under Approximate VCG Mechanisms

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read An immediate one-shot penalty can force truthful bidding in approximate VCG auctions, even under noisy monitoring, by exceeding the approximation-induced incentive gap scaled by detection probability.

desk verdict The paper's main theorem is not supported: Lemma 1 is false as stated, and the ε-tolerance in the detection model creates an irremovable gap that no penalty can close. read the letter →

arxiv 2512.00513 v2 pith:N6K4PPGL submitted 2025-11-29 cs.GT cs.MA

Xun Shao , Ryuuto Shimizu , Zhi Liu , Kaoru Ota , Mianxiong Dong This is my paper · ORCID
classification cs.GTcs.MA MSC 91B2691A18
keywords approximateVCGIoTenergytradingincentivecompatibilityimmediatepenaltysubgame-perfectequilibriumdoubleauctionmulti-agentreinforcementlearningtruthfulbidding
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 claims that a simple immediate penalty, applied in the same trading round, can restore truthful bidding in an α-approximate VCG double auction for IoT energy trading, despite approximate allocation and imperfect deviation detection. The enabling condition is that the penalty Π exceeds (1−α)C/ρ, where α is the allocation optimality ratio, C bounds a single agent's welfare impact, and ρ is the detection probability. If this holds, truthfulness becomes a subgame-perfect equilibrium in a single round, without reputation or repeated-game enforcement. The paper embeds the mechanism in a multi-agent reinforcement learning environment and reports that learned agents converge to truthful behavior precisely near this predicted threshold. A sympathetic reader would see this as a lightweight, identity-free tool for aligning autonomous prosumers with socially efficient outcomes in dynamic IoT markets.

What carries the argument

The load-bearing object is Lemma 1's bound on the approximation-induced incentive gap: for any agent and any misreport, the utility gain over truthful reporting is at most (1−α)C, where C is the maximum marginal contribution of any single agent to total welfare. This bound bridges the α-approximation guarantee of the allocation rule to a per-agent, per-round dollar figure that a one-shot penalty can offset. The penalty itself is intentionally myopic—no history, no identity, no continuation value—making the mechanism suitable for IoT environments with intermittent connectivity and noisy sensing.

What would settle it

For a small two-buyer/two-seller double auction with a chosen α-approximate allocation rule, exhaustively search over all possible misreports to compute the largest utility increase any agent can achieve, and compare that number with both (1−α)C and (1−α)W*. If the observed maximum gain exceeds (1−α)C, then the penalty threshold Π > (1−α)C/ρ does not guarantee truthful bidding.

Watch

Extended reading notes

Core claim

The central claim is Theorem 1: in an α-approximate VCG double auction with bounded marginal contribution C, deviation detection probability ρ, and immediate penalty Π, truthful reporting is a subgame-perfect equilibrium whenever Π > (1−α)C/ρ. The mechanism computes an approximate VCG allocation, applies a one-shot fine to any detected misreport, and deters deviation because the expected gain from misreporting is bounded by (1−α)C while the expected penalty is ρΠ. Since penalties do not affect future states, each round is an independent stage game, so enforcing truthfulness in the one-shot game extends to subgame perfection. The paper also reports experimental results in which the empiricall

Load-bearing premise

The proof that a single agent's misreport cannot gain more than (1−α)C, where C is the agent's own marginal welfare impact, rests on a one-sentence derivation; without additional structure on the approximate allocator, the true maximum gain could be larger, which would break the theorem's penalty threshold.

Editorial extensions

If this is right

  • A market operator can enforce truthful bids with a single fixed fine computed from three measurable quantities: allocation accuracy α, the per-agent welfare bound C, and detection reliability ρ.
  • Improving allocation accuracy or monitoring quality reduces the enforcement cost proportionally, giving operators a quantitative trade-off between computation and penalty size.
  • The same single-round enforcement logic applies to any approximate VCG setting with bounded marginal contributions, not only IoT energy trading.
  • Because the mechanism is myopic and identity-free, it is compatible with short-horizon or learning agents and with populations that change over time.
  • If the threshold is met, no rational agent of any type has a profitable deviation, so the market clears at near-efficient outcomes despite approximation and noise.

Reading between the lines

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

  • The paper's Lemma 1 bound may understate the true incentive gap: a direct VCG utility decomposition bounds the misreport gain by (1−α)W*, the total social-welfare loss, rather than the per-agent marginal contribution C. If C is small relative to W*, the required penalty could exceed the theorem's threshold, so the guarantee would need extra assumptions on the approximate allocator.
  • A testable extension is to measure the actual maximum misreport gain in the simulated environment and compare it against both (1−α)C and (1−α)W*; the better fit would indicate whether the enforcement condition should be revised.
  • The mechanism could be layered with adaptive penalties—for example, increasing Π when recent detections are low—to handle environments where ρ drifts over time.
  • The reported sharp phase transition near α≈0.7 offers a clean experimental falsification: repeating Plan A with independently estimated values of C and ρ should predict the exact boundary location, not just its qualitative shape.
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

3 major / 5 minor

Summary. The paper proposes an immediate-penalty enforcement mechanism for an alpha-approximate VCG double auction in IoT energy markets. It models deviation detection with tolerance epsilon and detection probability rho, and claims that a one-shot penalty Pi > (1-alpha)C/rho restores truthful reporting as a subgame-perfect equilibrium even when the allocation rule is only approximately welfare-maximizing and monitoring is noisy. The paper validates the mechanism experimentally in a multi-agent reinforcement learning environment, sweeping approximation ratio, tolerance, penalty, and discount factor, and reports that the empirically minimal penalty scales with (1-alpha) as predicted.

Significance. If the main theorem were correct, the paper would provide a simple and practically appealing way to enforce truthfulness in approximate VCG mechanisms with a single-round penalty, which is relevant to IoT energy trading. The paper has tangible strengths: a concrete IoT market model, a transparent closed-form enforcement condition, a MARL testbed with realistic load/solar traces, and attention to edge-computing latency. However, the central theoretical result is not established: the proof of the key incentive-gap lemma is missing, and the theorem is internally inconsistent with the paper's own epsilon-tolerance detection model. The empirical validation is also substantially circular because the approximate allocator is manufactured by scaling the exact optimal allocation. These issues are load-bearing, and I do not see a way to repair them within the current scope of the paper.

major comments (3)
  1. [Section IV-A / IV-D / Theorem 1] The detection rule in Section IV-A imposes the penalty only when |b_k - v'_k| > epsilon. Therefore, for any misreport with |b_k - v'_k| <= epsilon, D_k = 0 in Eq. (8), and the expected deviation gain in Eq. (5) reduces to Delta U_k = u_dev_k - u_truth_k. Lemma 1 itself allows this quantity to be strictly positive, up to (1-alpha)C > 0, and nothing in the lemma or the approximate-allocation model restricts profitable misreports to deviations larger than epsilon. Consequently, no finite Pi satisfying Eq. (7) excludes such deviations; the proof of Theorem 1 treats the rho Pi term as applying to 'every misreport', which contradicts the paper's own detection model. The theorem would need epsilon = 0 or a discretized type space with no deviations inside the epsilon interval, and neither assumption appears.
  2. [Section III-E, Lemma 1] The bound u_dev_k - u_truth_k <= (1-alpha)C is not derived. In an approximate VCG mechanism, the change in an agent's utility from a misreport is governed by the difference in the welfare expression evaluated at the two approximate allocations, including the externality/payment term that is based on the allocation without agent k. This difference is not bounded by the agent's own marginal contribution C to overall welfare. A direct expansion gives terms proportional to the total optimal welfare W*, which Section III-F itself says is typically much larger than C. As written, Lemma 1 is unsupported, and Theorem 1 inherits the problem. The authors need either a rigorous proof under explicit assumptions on A_alpha or a corrected threshold that depends on W* rather than only on C.
  3. [Section III-G / Section VI-C] The empirical confirmation is substantially circular. In Section III-G, A_alpha is not an independently defined approximation algorithm; it is constructed by scaling or thinning the exact optimal allocation so that W(x_alpha) is approximately alpha W*. The deviation incentives are therefore manufactured to scale with (1-alpha). Plan C then measures the minimal penalty Pi* and reports that it follows Pi* proportional to (1-alpha)C/rho; this largely reproduces the construction's input rather than testing the theory for a genuine approximate VCG allocation rule. The paper should replace this synthetic allocator with a real approximate solver (e.g., a greedy algorithm, LP rounding, or a heuristic with a formal approximation guarantee) and re-run the experiments.
minor comments (5)
  1. [Section IV-B] Typo: 'LoRaW AN' should be 'LoRaWAN'.
  2. [Equation (11)] The sign convention for eta_ch in the SoC update is unclear: a single efficiency factor cannot correctly model both charging and discharging losses. Please clarify or use separate charge/discharge efficiencies.
  3. [Section VI-A] The claim of a sharp phase transition near alpha ~ 0.7 depends on the fixed penalty Pi and the value of C, which are not specified in the figure description. Without this calibration, the comparison to the theoretical threshold is not verifiable.
  4. [General experimental presentation] Figures 2-5 are reported only qualitatively. Please provide error bars, the number of seeds, and, ideally, the trained policy code or a data repository so that the MARL results can be reproduced.
  5. [Theorem statement] The tolerance epsilon appears in the experiments but not in the enforcement condition (7). The paper should state explicitly how epsilon interacts with the theoretical guarantee, especially since the detection rule only penalizes deviations larger than epsilon.

Circularity Check

1 steps flagged · score 6.0 of 10

The empirical penalty map is partly wired into the allocator construction: A_α is made by scaling the exact optimum to W(x_α)≈αW*, so Plan C's finding that Π* scales with (1−α) largely reads back the construction's input rather than independently validating the theory.

  1. fitted input called prediction [Section III-G ('Instantiation of A_α for Simulation') and Section VI-C ('Plan C: Minimal Penalty Map Π*(α,ε)')]
    "We employ a synthetic two-step construction: 1) Compute a welfare-maximizing allocation x~ via convex optimization (small-scale oracle). 2) Derive x_α by scaling or thinning x~ such that W(x_α)≈αW(x~). This isolates the effect of approximation from confounding algorithmic details and allows MARL agents to experience controlled incentive distortions driven solely by (1−α). ... The empirical contours follow almost perfect linear decay in (1−α), matching the scaling Π⋆ ≈ (1−α)C/ρ."

    The empirical 'prediction' that the minimal penalty Π* is proportional to (1−α) is not an independent test of the α-approximate VCG theory. The paper deliberately constructs x_α so that the welfare ratio W(x_α)≈αW(x~) is forced by definition; hence the approximation-induced incentive distortion is, by construction, driven by (1−α). Plan C then estimates Π* and reports linear decay in (1−α), which is essentially recovering the same knob that was inserted into A_α. The experiment therefore confirms the construction's own input rather than validating the bound (1−α)C for a genuine approximate allocation algorithm with an independent error profile.

full rationale

The analytic derivation from Lemma 1 to Theorem 1 is algebraically straightforward and not circular: if the bound u_dev − u_truth ≤ (1−α)C were valid, then Π > (1−α)C/ρ would indeed make every deviation unprofitable. The circularity arises in the experimental confirmation. Section III-G defines A_α by scaling or thinning the exact welfare-maximizing allocation so that W(x_α)≈αW(x~), injecting the (1−α) dependence directly into the incentive structure. Section VI-C then presents the empirical linearity of Π* in (1−α) as evidence that the theory is correct, but that linearity is substantially predetermined by the allocator construction. A realistic approximate VCG allocator with an uncontrolled error profile could easily violate Lemma 1's per-agent C bound, since the construction's welfare loss is (1−α)W* rather than (1−α)C. I also note two non-circular correctness concerns that do not raise the circularity score: Lemma 1's proof is a single sentence and does not establish the claimed (1−α)C bound, and Section IV-A's ε-tolerance makes D_k=0 for small misreports while Theorem 1's proof applies the expected penalty to every misreport. These are logical gaps, not definitional reductions. No load-bearing self-citation chain appears in the paper.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The paper's central claim rests on the bounded-marginal-contribution assumption, an underspecified VCG payment formula, an unproved Lemma 1, and an ad-hoc detector that knows agents' true valuations. The simulation parameters (α, ε, ρ, Π, C) are chosen by hand; the synthetic allocator is constructed to satisfy W(x_α)≈αW*, making the empirical threshold comparison partly tautological.

free parameters (5)
  • C (bounded marginal contribution)
    Assumption 1 postulates |W(x)−W(x_{−k})|≤C for every agent and allocation, but C is never measured, computed from device data, or given a concrete value in the experiments; the main threshold (1−α)C/ρ depends on it.
  • Tolerance ε = 0.5–2.0
    Swept in Plan A and used to define when a deviation is detectable; values are chosen by hand and not derived from the named (but unlinked) IoT traces.
  • Detection probability ρ = 0.6–1.0 (discussion); 1.0 baseline
    Assigned representative values from generic IoT link conditions; no field measurement or model calibration is presented.
  • Approximation ratio α = 0.5–0.9
    Swept as a proxy for computation budget, but the allocator is constructed by scaling an exact solution to yield W(x_α)≈αW*, so α is a knob on a synthetic object, not a property of a real algorithm.
  • Penalty Π = Π0 baseline; ½Π0, 2Π0 varied
    Penalty strength is a design lever; the claimed minimal penalty Π* is the object of Plan C and is compared against the analytic threshold, but the environment is constructed to make that comparison match.
assumptions (5)
  • domain assumption Assumption 1: for any agent k and feasible allocation x, |W(x)−W(x_{−k})|≤C.
    Used in Lemma 1 to bound the incentive gap; the paper asserts this holds naturally for IoT devices, but does not bound W* or prove that misreport gains scale with C.
  • standard math The α-approximate allocator satisfies W(A_α(θ)) ≥ α W*(θ) for every report vector.
    Definition (3) of α-approximate VCG; standard, but the paper's experiments implement it by synthetic scaling rather than by a real algorithm.
  • domain assumption VCG payments take the same form with A_α replacing the exact optimizer, and the payment's h_i term is independent of the agent's own report.
    Section III-D says payments 'take the VCG form' but never specifies h_i; the proof of Lemma 1 depends on this structure without stating the payment formula.
  • ad hoc to paper The mechanism can compare a bid to the agent's true marginal valuation v'_k and detect deviations |b_k−v'_k|>ε with probability ρ.
    Section IV-A assumes access to the private type, which is exactly what a truthful mechanism is supposed to elicit; no procedure for obtaining v'_k is given. This detector is an invented capability, not a proven implementable component.
  • standard math Agent utilities are quasi-linear and valuations/costs are continuous, non-decreasing, concave.
    Section III-A; standard modeling assumptions for the VCG setting.
invented entities (1)
  • Deviation detector with access to true marginal valuation v'_k
    purpose: Detects misreports by checking |b_k−v'_k|>ε with probability ρ, enabling the immediate penalty.
    No design, estimator, or independent evidence is provided for how the market operator obtains each agent's true marginal valuation. This is a postulate that would be difficult to realize in a distributed IoT market where private types are the central information problem.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Truthful and Trustworthy IoT AI Agents via Immediate-Penalty Enforcement under Approximate VCG Mechanisms." pith.science (2026). https://pith.science/paper/N6K4PPGL

@misc{pith2026251200513,
  author       = {Pith},
  title        = {Pith review of: Truthful and Trustworthy IoT AI Agents via Immediate-Penalty Enforcement under Approximate VCG Mechanisms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N6K4PPGL}},
  note         = {Machine review of arXiv:2512.00513}
}
read the original abstract

The deployment of autonomous AI agents in Internet of Things (IoT) energy systems requires decision-making mechanisms that remain robust, efficient, and trustworthy under real-time constraints and imperfect monitoring. While reinforcement learning enables adaptive prosumer behaviors, ensuring economic consistency and preventing strategic manipulation remain open challenges, particularly when sensing noise or partial observability reduces the operator's ability to verify actions. This paper introduces a trust-enforcement framework for IoT energy trading that combines an approximate Vickrey-Clarke-Groves (VCG) double auction with an immediate one-shot penalty. Unlike reputation- or history-based approaches, the proposed mechanism restores truthful reporting within a single round, even when allocation accuracy is approximate and monitoring is noisy. We theoretically characterize the incentive gap induced by approximation and derive a penalty threshold that guarantees truthful bidding under bounded sensing errors. To evaluate learning-enabled prosumers, we embed the mechanism into a multi-agent reinforcement learning environment reflecting stochastic generation, dynamic loads, and heterogeneous trading opportunities. Experiments show that improved allocation accuracy reduces deviation incentives, the required penalty matches analytical predictions, and learned bidding behaviors remain stable and interpretable despite imperfect monitoring. These results demonstrate that lightweight penalty designs can reliably align strategic IoT agents with socially efficient energy-trading outcomes.

Figures

Figures reproduced from arXiv: 2512.00513 by the authors.

Figure 1
Figure 1. Overall architecture of the proposed α-approximate VCG double-auction mechanism with immediate penalty in a P2P IoT energy-trading environment. Autonomous prosumer agents map local observations to bids via MARL policies. The mechanism computes the approximate VCG allocation, detects significant misreports under noisy IoT sensing, and applies penalties. Resulting allocations and payments update the physical grid stat… view at source ↗
Figure 2
Figure 2. Truthfulness region in the (α, ε) plane (Plan A). Warm colors indicate high ε-truthful fraction. The sharp phase transition near α ≈ 0.7 matches the theoretical threshold at which (1 − α)C becomes small enough for the fixed penalty to dominate. Key findings. • Truthful regime (high α): For α ≥ 0.8, agents achieve TruthFracε > 0.9 regardless of ε. This aligns with the diminishing incentive gap (1 − α)C as α → 1. • St… view at source ↗
Figure 4
Figure 4. Minimal penalty threshold Π⋆(α, ε) (Plan C). Darker colors denote higher penalties. The monotonic decline with α and ε matches the theoretical scaling Π⋆ ∝ (1 − α)/ρ [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Robustness of truthful convergence under different RL hyper [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 2 linked inside Pith

  1. [1]

    The internet of things: A survey,

    L. Atzori, A. Iera, and G. Morabito, “The internet of things: A survey,” Computer Networks, vol. 54, no. 15, pp. 2787–2805, 2010

    2010

  2. [2]

    Edge computing: Vision and challenges,

    W. Shi, J. Cao, Q. Zhang, Y . Li, and L. Xu, “Edge computing: Vision and challenges,”IEEE Internet of Things Journal, vol. 3, no. 5, pp. 637–646, 2016

    2016

  3. [3]

    Internet of things: A survey on enabling technologies, protocols, and applications,

    A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and M. Ayyash, “Internet of things: A survey on enabling technologies, protocols, and applications,”IEEE Communications Surveys & Tutorials, vol. 17, no. 4, pp. 2347–2376, 2015

    2015

  4. [4]

    Prox- imal policy optimization algorithms,

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Prox- imal policy optimization algorithms,”arXiv preprint arXiv:1707.06347, 2017

    arXiv 2017

  5. [5]

    Multi- agent actor-critic for mixed cooperative–competitive environments,

    R. Lowe, Y . Wu, A. Tamar, J. Harb, P. Abbeel, and I. Mordatch, “Multi- agent actor-critic for mixed cooperative–competitive environments,” in Advances in Neural Information Processing Systems (NeurIPS), 2017, pp. 6379–6390

    2017

  6. [6]

    Security and privacy in the internet of things: Challenges and solutions,

    A. Alrawais, A. Alhothaily, C. Hu, and X. Shi, “Security and privacy in the internet of things: Challenges and solutions,”IEEE Internet of Things Journal, vol. 4, no. 5, pp. 1023–1035, 2017

    2017

  7. [7]

    Peer-to-peer trading in electricity networks: An overview,

    W. Tushar, B. Chai, C. Yuen, D. Smith, and H. V . Poor, “Peer-to-peer trading in electricity networks: An overview,”IEEE Transactions on Smart Grid, vol. 11, no. 4, pp. 3185–3200, 2020

    2020

  8. [8]

    Counterspeculation, auctions, and competitive sealed ten- ders,

    W. Vickrey, “Counterspeculation, auctions, and competitive sealed ten- ders,”Journal of Finance, vol. 16, no. 1, pp. 8–37, 1961

    1961

Show all 34 references
  1. [9]

    Multipart pricing of public goods,

    E. H. Clarke, “Multipart pricing of public goods,”Public Choice, vol. 11, pp. 17–33, 1971

    1971

  2. [10]

    Incentives in teams,

    T. Groves, “Incentives in teams,”Econometrica, vol. 41, no. 4, pp. 617– 631, 1973

    1973

  3. [11]

    Truth revelation in approximately efficient combinatorial auctions,

    D. Lehmann, L. I. O’Callaghan, and Y . Shoham, “Truth revelation in approximately efficient combinatorial auctions,”Journal of the ACM, vol. 49, no. 5, pp. 577–602, 2002

    2002

  4. [12]

    Nisan, T

    N. Nisan, T. Roughgarden, ´E. Tardos, and V . Vazirani,Algorithmic Game Theory. Cambridge University Press, 2007. IEEE INTERNET OF THINGS JOURNAL, VOL. XX, NO. XX, XXXX 2025 10

    2007

  5. [13]

    The folk theorem in repeated games with discounting or with incomplete information,

    D. Fudenberg and E. Maskin, “The folk theorem in repeated games with discounting or with incomplete information,”Econometrica, vol. 54, no. 3, pp. 533–554, 1986

    1986

  6. [14]

    G. J. Mailath and L. Samuelson,Repeated Games and Reputations: Long-Run Relationships. Oxford University Press, 2006

    2006

  7. [15]

    A comprehensive survey of multiagent reinforcement learning,

    L. Bus ¸oniu, R. Babuˇska, B. De Schutter, and D. Ernst, “A comprehensive survey of multiagent reinforcement learning,”IEEE Transactions on Systems, Man, and Cybernetics, Part C, vol. 38, no. 2, pp. 156–172, 2008

    2008

  8. [16]

    A survey and critique of multiagent deep reinforcement learning,

    R. Hern ´andez-Leal, B. Kartal, and M. E. Taylor, “A survey and critique of multiagent deep reinforcement learning,”Autonomous Agents and Multi-Agent Systems, vol. 33, no. 6, pp. 750–797, 2019

    2019

  9. [17]

    A practical guide to multi-agent reinforcement learning,

    Y . Yang, Y . Wen, and J. Wang, “A practical guide to multi-agent reinforcement learning,”Autonomous Agents and Multi-Agent Systems, vol. 35, no. 1, p. 13, 2021

    2021

  10. [18]

    Multi-agent reinforcement learning: A selective overview of theories and algorithms,

    K. Zhang, Z. Yang, and T. Bas ¸ar, “Multi-agent reinforcement learning: A selective overview of theories and algorithms,”Handbook of Rein- forcement Learning and Control, pp. 321–384, 2021

    2021

  11. [19]

    Secure cooperative edge computing for IoT: Reliable task offloading with imperfect information,

    Q. Xu, J. Ni, X. Cheng, J. Ren, and N. Kato, “Secure cooperative edge computing for IoT: Reliable task offloading with imperfect information,” IEEE Internet of Things Journal, vol. 7, no. 4, pp. 3113–3125, 2020

    2020

  12. [20]

    Data trustworthiness in IoT systems: A survey,

    Z. Yan, W. Ding, W. Feng, M. Xu, and N. Kato, “Data trustworthiness in IoT systems: A survey,”IEEE Internet of Things Journal, vol. 7, no. 5, pp. 4118–4135, 2020

    2020

  13. [21]

    Deep learning-based resource allocation for mobile edge computing in IoT networks,

    S. Sun, M. Peng, Y . Zhou, Y . Huang, and N. Kato, “Deep learning-based resource allocation for mobile edge computing in IoT networks,”IEEE Internet of Things Journal, vol. 7, no. 6, pp. 5534–5546, 2020

    2020

  14. [22]

    Reinforcement learning- based dynamic computation offloading in edge IoT systems,

    Y . Sun, X. Guan, J. Wu, Y . Ji, and N. Kato, “Reinforcement learning- based dynamic computation offloading in edge IoT systems,”IEEE Internet of Things Journal, vol. 6, no. 3, pp. 4664–4671, 2019

    2019

  15. [23]

    A dominant strategy double auction,

    R. P. McAfee, “A dominant strategy double auction,”Journal of Eco- nomic Theory, vol. 56, no. 2, pp. 434–450, 1992

    1992

  16. [24]

    Truthful mechanisms for one-parameter agents,

    A. Archer and ´E. Tardos, “Truthful mechanisms for one-parameter agents,” inFOCS, 2001, pp. 482–491

    2001

  17. [25]

    An analysis of ap- proximations for maximizing submodular set functions,

    G. L. Nemhauser, L. A. Wolsey, and M. Fisher, “An analysis of ap- proximations for maximizing submodular set functions,”Mathematical Programming, vol. 14, no. 1, pp. 265–294, 1978

    1978

  18. [26]

    Foundations of strategyproof mechanism design in ai,

    V . Conitzer and T. Sandholm, “Foundations of strategyproof mechanism design in ai,”AI Magazine, vol. 40, no. 2, pp. 45–58, 2019

    2019

  19. [27]

    AI for auctions: A survey,

    Y . Zheng, J. Zhang, and B. An, “AI for auctions: A survey,”Artificial Intelligence, vol. 301, p. 103576, 2022

    2022

  20. [28]

    Optimal auctions through deep learning,

    P. D ¨utting, Z. Feng, H. Narasimhan, D. Parkes, and S. Ravindranath, “Optimal auctions through deep learning,” inICML, 2019, pp. 1706– 1715

    2019

  21. [29]

    Reinforcement learning for incentive-compatible online auctions,

    G. Brero, B. Lubin, and S. Seuken, “Reinforcement learning for incentive-compatible online auctions,” inNeurIPS, vol. 32, 2019, pp. 3602–3612

    2019

  22. [30]

    Near-optimal auctions with ml agents,

    N. Golowich, H. Narasimhan, and D. Parkes, “Near-optimal auctions with ml agents,” inACM EC. ACM, 2021, pp. 372–392

    2021

  23. [31]

    Mechanism design for strategic ml agents,

    M. Curry, D. Schuurmans, K. Patil, R. Shariff, S. Schelter, and Z. Zhang, “Mechanism design for strategic ml agents,” inICML, vol. 162, 2022, pp. 4502–4523

    2022

  24. [32]

    Neural double auction mechanism,

    T. Suehara, K. Takeuchi, H. Kashima, S. Oyama, Y . Sakurai, and M. Yokoo, “Neural double auction mechanism,”arXiv preprint arXiv:2412.11465, 2024

    2024 arXiv

  25. [33]

    Deep reinforcement learning-based peer-to-peer energy trading for smart grids,

    X. Zhang, R. Zhang, and Q. Wu, “Deep reinforcement learning-based peer-to-peer energy trading for smart grids,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 52, no. 9, pp. 5592–5604, 2022. Xun Shao(Senior Member, IEEE) was a researcher with the National Ins...

    2022

  26. [2011]

    He is currently the Vice President and a Professor with the Muroran Institute of Technology, Muroran, Japan. He received numerous awards including the IEEE ComSoc Asia-Pacific Young Researcher Award (2017), Funai Research Award (2018), NISTEP Researcher Award (2018), Young Sci...

    2017

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.