acceptodds
Under review as a conference paper at ICLR 2027

FLAIRE: Zero-Trust and Rarity-Aware Client Incentivization for Heterogenous Federated Learning

Abstract

A challenge in Federated Learning is training under severe non-IID label skew. Incentivizing Rare-label clients is essential for maintaining global performance Existing incentive mechanisms often break down in practical deployments: they either assume honest client self-reporting of descriptors—exposing the system to strategic manipulation—or rely on server-side auxiliary datasets and external validators, violating core privacy and practical constraints. To resolve this dilemma, we propose FLAIRE (Federated Learning with Adaptive Incentives and Rewards Engine), an incentive architecture for heterogeneous FL built upon a zero-trust, zero-auxiliary-data, and zero-metadata foundation. FLAIRE achieves completely passive, metadata-free incentivization through two decoupled components: (1) a client profiling engine that infers local label distributions strictly via last-layer gradient inversion, requiring zero client-reported descriptors and zero server validation data; and (2) a label-rarity-aware incentive mechanism that assigns rewards inversely proportional to global label frequency. This design guarantees fair evaluation for rare-data providers ("Mavericks") while remaining resilient to free-riders and metadata tampering. FLAIRE is the first framework to unify passive, zero-trust profiling with rarity-aware contribution valuation in non-IID FL. Extensive evaluations on MNIST, Fashion-MNIST, EMNIST, and FEMNIST demonstrate that FLAIRE consistently boosts convergence speed and global model accuracy under extreme heterogeneity, achieving a minimum accuracy improvement of 5% across all settings and up to 10%–25% on the highly skewed benchmark scenarios tested.

Primary area:
other topics in machine learning (i.e., none of the above)
Keywords:
Federated Learning, Incentive Mechanisms, Data Heterogeneity, Non-IID, Label Skew, Zero-Trust Profiling, Collaborative Fairness, Label Rarity, Shapley Value Valuation, Maverick Clients

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_rwPNOpenReview ↗
Rating
1
Confidence
4
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

The paper proposes FLAIRE, an incentive framework for federated learning with heterogeneous client label distributions. It estimates label distributions from uploaded model updates using final-layer gradient inversion. The framework combines inverse-frequency label-rarity weights with a cosine-similarity-based approximation of Shapley contributions, normalizes contributions according to label coverage, and allocates rewards under a budget. Experiments on MNIST, Fashion-MNIST, EMNIST, and FEMNIST evaluate five mappings from cumulative contribution scores to client participation probabilities under non-overlapping label settings. Several configurations improve accuracy over fixed participation probabilities. However, the gains depend on the dataset and mapping, and some reported results contradict the claim of improvement across all settings.

Critical strengths

S1. The framework addresses a relevant limitation of contribution-based incentives. Clients holding rare but useful labels can be undervalued when reward allocation does not account for label scarcity or coverage. Combining contribution quality with rarity is a promising approach to encouraging their continued participation. This motivation connects client incentives to the global model’s coverage of underrepresented classes.

S2. The design reduces certain server-side data requirements. FLAIRE estimates client label distributions from existing model updates and approximates contributions without requiring a separate balanced server-side validation set. This reduces dependence on additional data and self-reported label information. The improvements reported in several configurations provide preliminary support for adapting participation to client contributions under strong label heterogeneity.

Critical weaknesses

W1. Inconsistent performance reporting undermines the central empirical claims. The paper claims at least a 5% improvement across all settings and consistent superiority of the Bayesian configuration. Table 3, however, reports an increase from 58.06% to 58.80% for FEMNIST A, or 0.74 percentage points, and a decrease from 66.28% to 61.44% for FEMNIST B, or 4.84 percentage points. These results contradict the stated generality of the gains, regardless of whether “5%” refers to an absolute or relative improvement. Several improvement entries also appear inconsistent with the corresponding accuracies. For FMNIST A, the Logistic configuration improves over the baseline by 4.67 percentage points, whereas the reported −20.59 appears to use the Bayesian configuration as its reference. For EMNIST A, the Linear configuration improves by 1.79 percentage points, while the reported improvement is 16.70%. These discrepancies require an explicit and consistently applied calculation convention. The concern is not that every configuration must outperform the baseline. It is that the reported evidence does not support the paper’s principal performance claim, and it is unclear which numerical entries are authoritative. Establishing reliable conclusions requires reconciling the tables and curves with the underlying experimental records and reporting variability across runs.

W2. Unresolved configurations and missing controls prevent reproducibility and attribution. The client-group counts sum to 24 for EMNIST Setting B and 22 for FEMNIST Setting B, while the main text states that there are 20 clients. This is consequential because client composition affects label-frequency weights, participation, and potentially local data allocation. Without identifying the configuration actually executed, the corresponding results cannot be reproduced or interpreted reliably. The paper also omits complete formulas and parameters for the five participation-probability mappings. Since these mappings directly determine participation, they are necessary to reproduce the comparisons. Moreover, comparing mapping functions against fixed participation does not isolate the effects of rarity weighting, coverage normalization, or inferred label profiles. The observed gains could depend primarily on the participation rule rather than the proposed combination of components. Taken together, these gaps extend beyond incomplete exposition. The current evidence does not establish which configuration produced the reported gains or which components account for them. Addressing this concern requires verified experimental specifications and controlled component comparisons.

Questions for authors

Q1. Numerical consistency. What do the parenthesized values in Tables 2 and 3 represent, and which configuration serves as the reference? Please reconcile the FMNIST A Logistic, EMNIST A Linear, and FEMNIST A Bayesian entries with their reported accuracies. Can the original per-round records and repeated-run summaries establish consistency between the tables and Figure 2?

Q2. Scope of the claims. How is the claim of “at least a 5% improvement across all settings” supported by the FEMNIST results? If the accuracies are correct, please revise the claim and discuss the Bayesian configuration’s degradation on FEMNIST B.

Q3. Experimental configuration. How many clients, and how many clients per group, were actually used in EMNIST and FEMNIST Setting B? Please provide the complete participation-probability update equations, parameter values, and configurations corresponding to each table and curve.

Q4. Attribution and profiling errors. With the training setup and participation mapping fixed, what happens when rarity weighting and coverage normalization are removed separately? How do reward allocations and training outcomes change when inferred label profiles are replaced with ground-truth profiles?

Reviewer_8QMJOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

In this work, the authors propose a federated learning incentive framework called FLAIRE (Federated Learning with Adaptive Incentives and Rewards Engine). FLAIRE consists primarily of two modules: a client profiling engine and a label-scarcity-aware incentive mechanism. FLAIRE encourages the participation of clients whose data is indispensable, yet often undervalued by existing methods, while simultaneously ensuring fairness. Experiments demonstrate the effectiveness of the proposed mechanism across various non-IID federated learning settings.

Critical strengths
  1. The paper addresses the issue that providers of rare labels may be undervalued in non-IID federated learning; it establishes a link between client incentives and label coverage, providing a clear motivation for the problem.

  2. Extensive experiments have been conducted to evaluate the performance of the proposed framework.

Critical weaknesses
  1. The title, abstract, and introduction repeatedly emphasize "zero-trust" and "without any trust assumptions," yet page 6 explicitly states: "We assume honest clients..."

  2. No empirical comparison with existing incentive mechanisms for federated learning was conducted.

  3. Not all of the configurations proposed by the authors achieved an improvement of at least 5%. For instance, see FEMNIST A in Table 3.

  4. There are several clear arithmetic errors in the experimental tables; some even alter the direction of an increase or decrease. For example, in Table 2: Logistic/FMNIST A, the value is −20.59%.

  5. The scaling and layout of the tables and figures are highly disorganized.

Questions for authors

This work is incomplete and unpolished.

Reviewer_1tigOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper proposes FLAIRE, a zero-trust, label-rarity-aware incentive framework for heterogeneous federated learning, to address the under-incentivization of clients holding rare labels and the reliance of existing incentive mechanisms on client-reported metadata or server-side auxiliary datasets. FLAIRE uses last-layer gradient inversion to infer clients' local label distributions and combines class-specific contribution evaluation with label rarity to allocate client rewards, aiming to improve the fairness of contribution evaluation for rare-data providers. Experiments under various non-IID settings report improvements in convergence speed and global model accuracy.

Critical strengths
  1. The paper addresses the potential underestimation of contributions from clients holding rare labels in heterogeneous federated learning. Incorporating label rarity into contribution evaluation and reward allocation is a relevant and practically meaningful research direction.

  2. The paper integrates gradient-based client profiling, class-specific contribution evaluation, and label-rarity-aware reward allocation into a unified framework without relying on client-reported label information or server-side auxiliary validation data. The overall approach is reasonably interpretable.

Critical weaknesses
  1. The discussion of related work is not sufficiently comprehensive, and the distinctive contributions of FLAIRE over existing contribution evaluation and incentive mechanisms are not clearly established. In particular, its differences from prior rarity-aware contribution valuation and client selection approaches remain insufficiently explained, and direct experimental comparisons with closely related methods are lacking.

  2. Some experimental results are inconsistent with the paper's main claims. Table 3 contains errors in the reported accuracy differences, and the Bayesian mapping performs worse than the baseline in one setting, contradicting the claims in the abstract and main text that accuracy improves by at least 5% across all settings. These inconsistencies raise concerns about the reliability of the reported conclusions.

Questions for authors
  1. Could the authors further clarify the limitations of closely related contribution evaluation and incentive mechanisms, explain the key differences and novel contributions of FLAIRE, and provide more targeted comparisons with existing approaches?

  2. How robust is the client profiling mechanism to errors in inferred label distributions, particularly the label-presence criterion in Eq. (3)? Is the contribution normalization in Eqs. (6)–(8) consistent with the monotonicity claim in Appendix Theorem 2? Could the authors provide ablation studies and analyses of reward allocation across clients with different label coverage to support the effectiveness of the proposed mechanism?

  3. Table 3 reports that the Bayesian mapping on FEMNIST A increases accuracy from 58.06% to 58.80%, but lists the difference as -0.74%; the Linear mapping on EMNIST A increases accuracy from 59.00% to 60.79%, but lists the difference as 16.70% rather than the actual 1.79 percentage points. Moreover, the Bayesian mapping on FEMNIST B decreases accuracy from 66.28% to 61.44%, contrary to the claim of at least a 5% improvement across all settings. Could the authors verify these results, explain the performance differences, and revise the corresponding claims?

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.