pith. sign in

arxiv: 2605.18020 · v1 · pith:IOLPLXHUnew · submitted 2026-05-18 · 💻 cs.LG

Federated Learning by Utility-Constrained Stochastic Aggregation for Improving Rational Participation

M Yashwanth , Arunabh Singh , Ashok Nayak , Sai Kiran Bulusu , Anirban Chakraborty This is my paper

Pith reviewed 2026-05-20 12:29 UTC · model grok-4.3

classification 💻 cs.LG
keywords federated learningrational participationclient retentionutility maximizationstochastic aggregationstatistical heterogeneity
0
0 comments X

The pith

FedUCA improves federated learning by using utility-constrained stochastic aggregation to sustain rational client participation

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

The paper establishes that standard federated learning assumes clients will always share updates, yet in real cross-silo settings with heterogeneous data rational clients may withdraw if collaboration fails to improve their local model performance above a personal threshold. FedUCA counters this by recasting the server as an optimizer that performs stochastic aggregation under explicit utility constraints designed to meet those thresholds while still advancing the global model. Experiments on standard datasets show the approach yields higher client retention and stronger final models than conventional methods. A sympathetic reader would care because unchecked attrition can degrade the shared model or cause the entire training process to collapse, leaving participants worse off than if they had trained independently.

Core claim

FedUCA formalizes the server's role as an optimizer that maximizes global model performance by sustaining client participation, implemented through utility-constrained stochastic aggregation that adjusts the process to keep each client's perceived local benefit above its dropout threshold in statistically heterogeneous environments.

What carries the argument

Utility-Constrained Stochastic Aggregation: the server's mechanism for selecting aggregation weights or participants stochastically while enforcing constraints derived from modeled client local performance thresholds to encourage continued involvement.

If this is right

  • Client retention rates rise in heterogeneous cross-silo settings.
  • Global model performance improves from the sustained diversity of participating clients.
  • The risk of federated training collapsing due to mass withdrawals decreases.
  • Server aggregation decisions become a direct lever for shaping client participation behavior.

Where Pith is reading between the lines

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

  • The same utility-modeling approach could be adapted to design explicit participation incentives in other decentralized training systems.
  • Explicitly tracking client rationality may prove necessary when scaling federated methods to production environments with independent actors.
  • Future work could test whether relaxing the rationality assumption still preserves the retention gains observed here.

Load-bearing premise

Clients behave as rational utility maximizers whose local performance thresholds can be accurately modeled and influenced by the server's aggregation choices.

What would settle it

A controlled experiment that records actual client opt-out decisions across rounds and checks whether the observed dropout rates align with predictions from the utility model under FedUCA versus standard aggregation.

Figures

Figures reproduced from arXiv: 2605.18020 by Anirban Chakraborty, Arunabh Singh, Ashok Nayak, M Yashwanth, Sai Kiran Bulusu.

Figure 1
Figure 1. Figure 1: Comparison of participation rates across different [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Impact of Dirichlet concentration δ on participation for CIFAR-10 [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Representative empirical loss and utility profiles across two different clients, rounds, on the [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Representative empirical loss and utility profiles across two different clients, rounds, on the [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Data distribution heatmap for CIFAR-10 (0.5, 1.0) showing Training and Validation splits [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Data distribution heatmap for CIFAR-10 (1.0, 1.0) showing Training and Validation splits [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Empirical aggregation discrepancy ∥1/N −αt∥ 2 2 across communication rounds for FedUCA with ns = 5 on (a) CIFAR-10 (N = 10) and (b) CIFAR-100 (N = 20). The dashed line marks the theoretical worst-case bound (ns − 1)/N from Lemma A.10. Empirical values remain well below the bound throughout training. The gradual increase reflects growing heterogeneity in fitted client utilities ζk as local models diverge, w… view at source ↗
read the original abstract

Federated Learning (FL) algorithms implicitly assume that clients passively comply with server-side orchestration by sharing local model updates upon server request. However, this overlooks an important aspect in real-world cross-silo environments: clients are often rational agents who may prioritize their utilities such as local model performance over that of the global model. In settings with significant statistical heterogeneity, rational clients may opt out of the federation if the perceived benefits of collaboration fail to meet their local utility thresholds. Such attrition degrades the global model performance and can lead to the collapse of the federated training process. In this work, we introduce FedUCA, (Federated Learning by Utility-Constrained Stochastic Aggregation for Improving Rational Participation), a framework that formalizes the server's role as an optimizer seeking to maximize global model performance by sustaining client participation. We substantiate our framework through extensive experiments on standard datasets demonstrating that by prioritizing participation feasibility, FedUCA achieves significantly higher client retention and, consequently, a superior global model performance.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces FedUCA, a framework that models clients in federated learning as rational utility maximizers who may drop out if local performance thresholds are not met. It proposes utility-constrained stochastic aggregation at the server to maximize global model performance by sustaining participation, and reports experiments on standard datasets showing higher client retention and improved global accuracy as a result.

Significance. If the causal attribution to participation holds, the work addresses a practically relevant gap in cross-silo FL where statistical heterogeneity can cause rational attrition and training collapse. The emphasis on explicit participation modeling and the reported retention gains would be a useful contribution to incentive-aware FL literature.

major comments (2)
  1. [§4] §4 (Experiments) and the associated figures: the central claim that superior global performance is 'consequently' due to higher client retention is not isolated from the change in aggregation operator. No ablation is reported that holds the stochastic aggregation rule fixed while varying only the utility constraint (or vice versa), leaving open the possibility that performance differences arise from altered update statistics rather than retention.
  2. [§3] §3 (Framework formulation): the optimization objective that trades off global loss against participation feasibility is stated at a high level, but the precise functional form of the utility constraint and how it enters the stochastic aggregation weights is not derived or shown to be parameter-free; this makes it difficult to assess whether the reported improvements are robust to modeling choices for client utilities.
minor comments (2)
  1. [Abstract] Abstract and §1: the claim of 'extensive experiments' is not accompanied by details on number of clients, heterogeneity levels, or statistical significance of retention and accuracy differences; these should be summarized early for clarity.
  2. [§2] Notation in §2: the definition of client utility threshold and its relation to local model performance should be made explicit with an equation, as the current prose description leaves the mapping from model quality to participation decision ambiguous.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment below and have revised the manuscript to strengthen the presentation and empirical support where appropriate.

read point-by-point responses

  1. Referee: [§4] §4 (Experiments) and the associated figures: the central claim that superior global performance is 'consequently' due to higher client retention is not isolated from the change in aggregation operator. No ablation is reported that holds the stochastic aggregation rule fixed while varying only the utility constraint (or vice versa), leaving open the possibility that performance differences arise from altered update statistics rather than retention.

    Authors: We agree that an explicit ablation isolating the utility constraint from the stochastic aggregation operator would strengthen the causal attribution. In the revised manuscript we have added such an ablation: we fix the stochastic aggregation rule and vary only the utility-constraint threshold, reporting the resulting changes in retention and global accuracy. This new experiment supports that the observed gains are driven by sustained participation rather than solely by altered update statistics. revision: yes

  2. Referee: [§3] §3 (Framework formulation): the optimization objective that trades off global loss against participation feasibility is stated at a high level, but the precise functional form of the utility constraint and how it enters the stochastic aggregation weights is not derived or shown to be parameter-free; this makes it difficult to assess whether the reported improvements are robust to modeling choices for client utilities.

    Authors: We have expanded the derivation in Section 3. The utility constraint is expressed as a per-client feasibility condition on expected local utility gain; this condition directly modulates the sampling probabilities inside the stochastic aggregation weights via a normalized, closed-form reweighting that uses only quantities already computed during local training. The resulting procedure introduces no additional free parameters beyond the original FL hyperparameters. We also added a short sensitivity study confirming robustness to reasonable variations in the client utility model. revision: yes

Circularity Check

0 steps flagged

No circularity detected; framework description lacks explicit derivations or equations that reduce to inputs by construction.

full rationale

The provided abstract and description introduce FedUCA as a framework that formalizes the server as an optimizer maximizing global performance by sustaining participation, substantiated via experiments on standard datasets showing higher retention and superior performance. No equations, optimization formulations, self-citations, or ansatzes are quoted or described that would allow reduction of any prediction or result to fitted parameters or prior self-referential definitions. The central claim rests on empirical outcomes rather than a closed mathematical chain, making the derivation self-contained against external benchmarks with no load-bearing self-referential steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; all assessments rest on high-level claims only.

pith-pipeline@v0.9.0 · 5713 in / 1048 out tokens · 44704 ms · 2026-05-20T12:29:26.906668+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages · 1 internal anchor

  1. [1]

    Federated learning based on dynamic regularization

    Durmus Alp Emre Acar, Yue Zhao, Ramon Matas, Matthew Mattina, Paul Whatmough, and Venkatesh Saligrama. Federated learning based on dynamic regularization. InInternational Conference on Learning Representations, 2021. URL https://openreview.net/forum? id=B7v4QMR6Z9w

    work page 2021

  2. [2]

    A rewriting system for convex optimization problems.Journal of Control and Decision, 5(1):42–60, 2018

    Akshay Agrawal, Robin Verschueren, Steven Diamond, and Stephen Boyd. A rewriting system for convex optimization problems.Journal of Control and Decision, 5(1):42–60, 2018

    work page 2018

  3. [3]

    Aliprantis and K.C

    C.D. Aliprantis and K.C. Border.Infinite Dimensional Analysis: A Hitchhiker’s Guide. Studies in Economic Theory. Springer, 1999. ISBN 9783540658542. URL https://books.google. co.in/books?id=6jjY2Vi3aDEC

    work page 1999

  4. [4]

    Towards more sustainable enterprise data and application management with cross silo federated learning and analytics.arXiv preprint arXiv:2312.14628, 2023

    Hongliu Cao. Towards more sustainable enterprise data and application management with cross silo federated learning and analytics.arXiv preprint arXiv:2312.14628, 2023

    work page arXiv 2023

  5. [5]

    To federate or not to federate: Incentivizing client participation in federated learning

    Yae Jee Cho, Divyansh Jhunjhunwala, Tian Li, Virginia Smith, and Gauri Joshi. To federate or not to federate: Incentivizing client participation in federated learning. InWorkshop on Federated Learning: Recent Advances and New Challenges (in Conjunction with NeurIPS 2022), 2022. URLhttps://openreview.net/forum?id=pG08eM0CQba

    work page 2022

  6. [6]

    Fed- biomed: Open, transparent and trusted federated learning for real-world healthcare applications

    Francesco Cremonesi, Marc Vesin, Sergen Cansiz, Yannick Bouillard, Irene Balelli, Lucia Innocenti, Riccardo Taiello, Santiago Silva, Samy-Safwan Ayed, Melek Önen, et al. Fed- biomed: Open, transparent and trusted federated learning for real-world healthcare applications. 10 InFederated Learning Systems: Towards Privacy-Preserving Distributed AI, pages 19–...

    work page 2025

  7. [7]

    Federated learning for predicting clinical outcomes in patients with covid-19.Nature medicine, 27(10): 1735–1743, 2021

    Ittai Dayan, Holger R Roth, Aoxiao Zhong, Ahmed Harouni, Amilcare Gentili, Anas Z Abidin, Andrew Liu, Anthony Beardsworth Costa, Bradford J Wood, Chien-Sung Tsai, et al. Federated learning for predicting clinical outcomes in patients with covid-19.Nature medicine, 27(10): 1735–1743, 2021

    work page 2021

  8. [8]

    Imagenet: A large- scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large- scale hierarchical image database. In2009 IEEE Conference on Computer Vision and Pattern Recognition, pages 248–255, 2009. doi: 10.1109/CVPR.2009.5206848

    work page doi:10.1109/cvpr.2009.5206848 2009

  9. [9]

    CVXPY: A Python-embedded modeling language for convex optimization.Journal of Machine Learning Research, 17(83):1–5, 2016

    Steven Diamond and Stephen Boyd. CVXPY: A Python-embedded modeling language for convex optimization.Journal of Machine Learning Research, 17(83):1–5, 2016

    work page 2016

  10. [10]

    Loss surfaces, mode connectivity, and fast ensembling of dnns.Advances in neural information processing systems, 31, 2018

    Timur Garipov, Pavel Izmailov, Dmitrii Podoprikhin, Dmitry P Vetrov, and Andrew G Wilson. Loss surfaces, mode connectivity, and fast ensembling of dnns.Advances in neural information processing systems, 31, 2018

    work page 2018

  11. [11]

    Data shapley: Equitable valuation of data for machine learning

    Amirata Ghorbani and James Zou. Data shapley: Equitable valuation of data for machine learning. InInternational Conference on Machine Learning (ICML), 2019

    work page 2019

  12. [12]

    Pursuing overall welfare in federated learning through sequential decision making, 2024

    Seok-Ju Hahn, Gi-Soo Kim, and Junghye Lee. Pursuing overall welfare in federated learning through sequential decision making, 2024. URLhttps://arxiv.org/abs/2405.20821

    work page arXiv 2024

  13. [13]

    Federated Learning for Mobile Keyboard Prediction

    Andrew Hard, Kanishka Rao, Rajiv Mathews, Swaroop Ramaswamy, Françoise Beaufays, Sean Augenstein, Hubert Eichner, Chloé Kiddon, and Daniel Ramage. Federated learning for mobile keyboard prediction, 2019. URLhttps://arxiv.org/abs/1811.03604

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  14. [14]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

    work page 2016

  15. [15]

    Fedvarp: Tack- ling the variance due to partial client participation in federated learning

    Divyansh Jhunjhunwala, Pranay Sharma, Aushim Nagarkatti, and Gauri Joshi. Fedvarp: Tack- ling the variance due to partial client participation in federated learning. InUncertainty in Artificial Intelligence, pages 906–916. PMLR, 2022

    work page 2022

  16. [16]

    Incentive mechanism for reliable federated learning: A joint optimization approach to combining reputation and contract theory

    Jiawen Kang, Zehui Xiong, Dusit Niyato, et al. Incentive mechanism for reliable federated learning: A joint optimization approach to combining reputation and contract theory. InIEEE Internet of Things Journal, 2019

    work page 2019

  17. [17]

    Scaffold: Stochastic controlled averaging for federated learning

    Sai Praneeth Karimireddy, Satyen Kale, Mehryar Mohri, Sashank Reddi, Sebastian Stich, and Ananda Theertha Suresh. Scaffold: Stochastic controlled averaging for federated learning. In International conference on machine learning, pages 5132–5143. PMLR, 2020

    work page 2020

  18. [18]

    Mechanisms that incentivize data sharing in federated learning

    Sai Praneeth Karimireddy, Wenshuo Guo, and Michael Jordan. Mechanisms that incentivize data sharing in federated learning. InWorkshop on Federated Learning: Recent Advances and New Challenges (in Conjunction with NeurIPS 2022), 2022. URL https://openreview. net/forum?id=Bx4Sz-N5K3J

    work page 2022

  19. [19]

    Understanding black-box predictions via influence functions

    Pang Wei Koh and Percy Liang. Understanding black-box predictions via influence functions. InInternational conference on machine learning, pages 1885–1894. PMLR, 2017

    work page 2017

  20. [20]

    Incentivizing federated learning.arXiv preprint arXiv:2205.10951, 2022

    Shuyu Kong, You Li, and Hai Zhou. Incentivizing federated learning.arXiv preprint arXiv:2205.10951, 2022

    work page arXiv 2022

  21. [21]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009

    work page 2009

  22. [22]

    Visualizing the loss landscape of neural nets.Advances in neural information processing systems, 31, 2018

    Hao Li, Zheng Xu, Gavin Taylor, Christoph Studer, and Tom Goldstein. Visualizing the loss landscape of neural nets.Advances in neural information processing systems, 31, 2018

    work page 2018

  23. [23]

    Fair resource allocation in federated learning.International Conference on Learning Representations (ICLR), 2020

    Tian Li, Anit Kumar Sahu, Ameet Talwalkar, and Virginia Smith. Fair resource allocation in federated learning.International Conference on Learning Representations (ICLR), 2020. 11

    work page 2020

  24. [24]

    Federated optimization in heterogeneous networks.Proceedings of Machine learning and systems, 2:429–450, 2020

    Tian Li, Anit Kumar Sahu, Manzil Zaheer, Maziar Sanjabi, Ameet Talwalkar, and Virginia Smith. Federated optimization in heterogeneous networks.Proceedings of Machine learning and systems, 2:429–450, 2020

    work page 2020

  25. [25]

    Incentive mechanism design for unbiased federated learning with randomized client participation

    Bing Luo, Yutong Feng, Shiqiang Wang, Jianwei Huang, and Leandros Tassiulas. Incentive mechanism design for unbiased federated learning with randomized client participation. In 2023 IEEE 43rd International Conference on Distributed Computing Systems (ICDCS), pages 545–555. IEEE, 2023

    work page 2023

  26. [26]

    Communication-efficient learning of deep networks from decentralized data

    Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas. Communication-efficient learning of deep networks from decentralized data. InArtificial intelligence and statistics, pages 1273–1282. PMLR, 2017

    work page 2017

  27. [27]

    Mitchell

    T. Mitchell. Twenty Newsgroups, 1997. URL https://archive.ics.uci.edu/ml/ datasets/Twenty+Newsgroups

    work page 1997

  28. [28]

    Roger B. Myerson. Optimal auction design.Mathematics of Operations Research, 6(1):58–73,

    work page

  29. [29]

    URLhttp://www.jstor.org/stable/3689266

    ISSN 0364765X, 15265471. URLhttp://www.jstor.org/stable/3689266

    work page arXiv

  30. [30]

    Federated learning with buffered asynchronous aggregation

    John Nguyen, Kshitiz Malik, Hongyuan Zhan, Ashkan Yousefpour, Mike Rabbat, Mani Malek, and Dzmitry Huba. Federated learning with buffered asynchronous aggregation. InInternational conference on artificial intelligence and statistics, pages 3581–3607. PMLR, 2022

    work page 2022

  31. [31]

    Valentin V . Petrov. On lower bounds for tail probabilities.Journal of Statistical Plan- ning and Inference, 137(8):2703–2705, 2007. ISSN 0378-3758. doi: https://doi.org/ 10.1016/j.jspi.2006.02.015. URL https://www.sciencedirect.com/science/article/ pii/S0378375807000213. 5th St. Petersburg Workshop on Simulation

    work page doi:10.1016/j.jspi.2006.02.015 2007

  32. [32]

    John W. Pratt. Risk aversion in the small and in the large.Econometrica, 32(1/2):122–136,

    work page

  33. [33]

    URLhttp://www.jstor.org/stable/1913738

    ISSN 00129682, 14680262. URLhttp://www.jstor.org/stable/1913738

    work page arXiv

  34. [34]

    Motivating workers in federated learning: A stackelberg game perspective, 2019

    Yunus Sarikaya and Ozgur Ercetin. Motivating workers in federated learning: A stackelberg game perspective, 2019. URLhttps://arxiv.org/abs/1908.03092

    work page arXiv 2019

  35. [35]

    Schmitz.Journal of Institutional and Theoretical Economics (JITE) / Zeitschrift für die gesamte Staatswissenschaft, 162(3):535–540, 2006

    Patrick W. Schmitz.Journal of Institutional and Theoretical Economics (JITE) / Zeitschrift für die gesamte Staatswissenschaft, 162(3):535–540, 2006. ISSN 09324569. URL http: //www.jstor.org/stable/40752600

    work page arXiv 2006

  36. [36]

    A principled approach to data valuation for federated learning

    Tianhao Wang, Johannes Rausch, Ce Zhang, Ruoxi Jia, and Dawn Song. A principled approach to data valuation for federated learning. pages 153–167, 2020

    work page 2020

  37. [37]

    Incentive mechanism for federated learning with random client selection.IEEE Transactions on Network Science and Engineering, 11(2):1922–1933, 2024

    Hongyi Wu, Xiaoying Tang, Ying-Jun Angela Zhang, and Lin Gao. Incentive mechanism for federated learning with random client selection.IEEE Transactions on Network Science and Engineering, 11(2):1922–1933, 2024. doi: 10.1109/TNSE.2023.3334476

    work page doi:10.1109/tnse.2023.3334476 1922

  38. [38]

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms, 2017

    Han Xiao, Kashif Rasul, and Roland V ollgraf. Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms, 2017. URL https://arxiv.org/abs/1708. 07747

    work page 2017

  39. [39]

    Daringfed: a dynamic bayesian persuasion pricing for online federated learning under two-sided incomplete information

    Yun Xin, Jianfeng Lu, Shuqin Cao, Gang Li, Haozhao Wang, and Guanghui Wen. Daringfed: a dynamic bayesian persuasion pricing for online federated learning under two-sided incomplete information. InProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, pages 6687–6695, 2025

    work page 2025

  40. [40]

    Contract-based incentive mechanism for federated learning in edge computing system

    Lu Yu, Zheng Chang, and Zhiwei Zhao. Contract-based incentive mechanism for federated learning in edge computing system. In2024 IEEE Wireless Communications and Networking Conference (WCNC), pages 1–6, 2024. doi: 10.1109/WCNC57260.2024.10570831

    work page doi:10.1109/wcnc57260.2024.10570831 2024

  41. [41]

    A comprehensive survey of incentive mechanism for federated learning, 2021

    Rongfei Zeng, Chao Zeng, Xingwei Wang, Bo Li, and Xiaowen Chu. A comprehensive survey of incentive mechanism for federated learning, 2021. URLhttps://arxiv.org/abs/2106. 15406

    work page 2021

  42. [42]

    self-interested

    Ning Zhang, Xiaoqing Xu, Liuyihui Qian, Xiaojun Liu, Juan Wu, and Hong Tang. Auction- based incentive mechanism in federated learning considering communication path finding.IEEE Access, 12:139336–139345, 2024. doi: 10.1109/ACCESS.2024.3425948. 12 A Technical appendices and supplementary material A.1 Notation Table 4: Summary of notation used in the paper....

    work page doi:10.1109/access.2024.3425948 2024

  43. [43]

    NX k=1 αt kI t k ht k − ∇fk(θ(t,0)) 2 # | {z } F1 +3E

    using the inequality||A|| 2 2 ≤ ||A||1||A||∞: •||R T ||1 = maxj P i Ri,j = max(row sums ofR) = 1. •||R T ||∞ = maxi P j Ri,j = max(col sums ofR) = ns N . Therefore, foranyfeasible optimization outcome, the spectral norm is deterministically bounded by ||RT ||2 2 ≤ ns N . Taking the expectation over the Dirichlet distributions: Es|| 1 N 1N −α t||2 2 ≤ ns N...

    work page