Pith. sign in

REVIEW 2 major objections 6 minor 1 cited by

Universal Refinement without Interaction: Order-Optimal 1-Bit Mean Estimation

T0 review · 2 major / 6 minor · reviewed 2026-07-31 · grok-4.5

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

Pith's one-line read Interaction is unnecessary: fully non-adaptive 1-bit queries match adaptive order-optimal rates for mean estimation under finite central moments.

desk verdict Clean affirmative answer to the Lau–Scarlett non-adaptive open problem via decoder-side universal refinement; the math holds as a reduction on top of their localization block. read the letter →

arxiv 2607.24358 v1 pith:MICIARZ7 submitted 2026-07-27 cs.IT math.IT

Yuchen Miao This is my paper · ORCID
classification cs.ITmath.IT MSC 62G0594A2968Q32
keywords 1-bitmeanestimationnon-adaptiveprotocolsfinitecentralmomentsdistributedpublic-coindecoder-siderefinementperiodicquantization
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

When each device can send only one bit about a fresh sample, adaptive threshold queries were known to achieve the best possible sample rates under a single central-moment bound. This paper shows those same rates are attainable with no interaction at all, provided the bits may answer arbitrary measurable questions fixed in advance. A public-coin protocol generates every localization and refinement query in one batch; after a coarse center is decoded, it changes only how the already-stored refinement bits are weighted, never which sets were queried. Two constructions—dyadic safe periodic residues and continuous shifted random grids—deliver the standard three refinement regimes (light, critical, and heavy tails) plus an additive localization cost. In the small-error, high-confidence range where matching lower bounds already exist, the resulting sample complexity is minimax optimal.

What carries the argument

Universal decoder-side refinement: all measurable 1-bit queries are fixed before any message is seen; a later-decoded coarse center only reinterprets stored bits. The dyadic construction uses safe half-period residues, correlated thresholding, and an adjacent-scale telescope whose variance is supported only on boundary crossings; the continuous construction integrates a compactly supported kernel over random grid widths with adjacent-cell gradients. Moment-matched scale sampling yields the three optimal tail regimes.

What would settle it

Exhibit a matching lower bound, or a concrete distribution family in D(k,λ,σ), showing that every fully non-adaptive protocol with arbitrary measurable 1-bit queries still needs asymptotically more samples than the stated rate r_k in the small-error, high-confidence regime where adaptive lower bounds already apply.

Watch

Extended reading notes

Core claim

For every fixed moment order k>1, a fully non-adaptive public-coin 1-bit protocol is (ε,δ)-accurate over the class of distributions with mean bounded by λ and k-th central moment bounded by σ^k, using a sample size of the same order as the best adaptive protocols: an additive localization term 1+log(λ/σ) plus the usual refinement cost in σ/ε and log(1/δ) that depends on whether k is above, equal to, or below 2.

Load-bearing premise

The argument needs arbitrary measurable query sets (generally nonlocal unions of intervals) and shared public randomness, and it sits on top of an existing non-adaptive localization block; it does not claim the same rates for ordinary threshold or interval queries alone.

Editorial extensions

If this is right

  • Zero adaptive rounds suffice for order-optimal scalar 1-bit mean estimation once arbitrary measurable queries are allowed.
  • All localization and refinement bits can be issued in a single parallel batch, removing sequential latency.
  • The known adaptive–non-adaptive gap for this task is an artifact of restricting to threshold or interval queries, not of the 1-bit budget itself.
  • In the parameter range of existing high-confidence lower bounds, the non-adaptive sample complexity is minimax optimal up to k-dependent constants.

Reading between the lines

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

  • The same separate-query-from-decoder pattern may transfer to other one-dimensional distributed tasks where a coarse location is easy to code but fine residuals are location-dependent.
  • Because both a discrete telescope and a continuous integral identity work, the phenomenon is about decoder-side recentering rather than one algebraic representation.
  • Extending the idea past the scalar setting will need new geometry: coordinate-wise application need not preserve optimal dimension dependence, as the paper notes remains open.
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.

Referee Report

2 major / 6 minor

Summary. The paper addresses the Lau–Scarlett open problem: whether fully non-adaptive arbitrary 1-bit queries can match the adaptive minimax rates for mean estimation over the class D(k,λ,σ) with |EX|≤λ and k-th central moment ≤σ^k, k>1 fixed. The main result (Theorem 2.1) constructs a fully non-adaptive public-coin protocol whose sample complexity matches the adaptive rate r_k(λ,σ,ε,δ) in (1), up to k-dependent constants, in the small-error regime ε≤c_kσ. The protocol combines an imported non-adaptive localization block (Proposition 2.2, Lau–Scarlett Theorem 16) with two universal refinement constructions: a dyadic scheme using safe periodic residues with an adjacent-scale telescope and moment-matched scale sampling p_j ∝ L_j^{(2−k)/2}, and a continuous-scale scheme using shifted random grids, Rademacher cell coloring, and a compactly supported kernel identity. All queries are fixed before communication; the decoded center affects only decoder-side weights. I verified the internal derivations in detail (safe-phase lemma, correlated-threshold identity, telescope and terminal bias (20)–(22), the three variance regimes (23), the k=2 pointwise summation (24), median-of-means concentration, and the continuous-scale analogues (48)–(59)) and found them correct.

Significance. If the imported localization theorem holds as quoted, this resolves a named COLT open problem and establishes a sharp separation: interaction is unnecessary for scalar finite-moment 1-bit mean estimation with arbitrary measurable queries, while it is provably necessary under threshold/interval restrictions. Strengths worth naming: two complete and internally verified constructions with full proofs (Appendices A–B), including a clean moment-matched scale allocation that yields all three tail regimes from one importance distribution; a careful k=2 boundary analysis that avoids an artificial extra logarithm via the pointwise bound (24), with honest disclosure of non-uniformity as k→2 (Remark 4.3); matching against an external minimax lower bound rather than a fitted benchmark; and a reproducible numerical appendix with Rao–Blackwellized evaluation, mechanism tests, and a public artifact.

major comments (2)
  1. [§2, Proposition 2.2] The entire refinement analysis is conditioned on the localization guarantee: Lemma 2.3, Eq. (3), and the moment inflation in (17) all begin from Proposition 2.2, which imports Theorem 16 of Lau–Scarlett (2026b, version 2). If that theorem's guarantee differs in form — expected rather than high-probability interval length, an adaptive or private-coin model, a stronger moment premise, or a different confidence dependence — the conditional decoupling and the sample-complexity accounting in §A.6 do not go through as stated. Because this is the single point whose failure would collapse Theorem 2.1, I ask the authors to (i) restate the imported theorem in full (hypotheses, model, guarantee, confidence scaling) in an appendix, (ii) verify explicitly that each hypothesis holds on D(k,λ,σ) (the Lyapunov check in (2) covers only the first-moment premise), and (iii) pin the arXiv version number in
  2. [§4.4, Corollary 4.4] The corollary asserts minimax optimality of the fully non-adaptive class by combining Theorem 2.1 with Lau–Scarlett's Theorem 9. The argument takes c'_k = min{cup_k, clb_k} and δ < δ_0, which is fine, but it should also verify that the lower bound's localization term Ω(log(λ/σ)) is proved under the same premise (|EX| ≤ λ, k-th central moment ≤ σ^k) and the same query model, and that the refinement lower bounds hold for arbitrary measurable queries rather than only thresholds. A two-sentence verification would close this; as written the reader must trust that the ranges and models align.
minor comments (6)
  1. [Title page] The affiliation/email line and the repository link run together ('miaoyc@mails.neu.edu.cn /githubhttps://...'); please repair the header formatting.
  2. [§1.3, second paragraph] The notation R(B − B_c) for the correlated-threshold statistic is used before it is formally introduced in Lemma 3.2; a forward pointer would help first-time readers.
  3. [§5, Eq. (34) vs. Appendix B.3, Eq. (58)] The variance bound is stated with ℓ(τ/ϵ) in Proposition 5.1 (34) but proved as Cτ²log(eτ/ϵ) in Lemma B.3; since ℓ(x) = 1 + log x, these agree only up to the convention that the log is at least one — please use one notation or note the equivalence explicitly.
  4. [Appendix C.1] The experiments supply the decoder with an oracle center c = 0, so only the refinement block is validated; the localization block and the end-to-end pipeline are not exercised. This is reasonable but should be stated in one sentence in the main text when the experiments are mentioned.
  5. [Appendix C, Figures 3–4] The ordinate of Figure 3 is defined only via v_k in Appendix C.3; adding the definition to the caption would make the figure self-contained. The color-independent hatching in Figure 4 is appreciated.
  6. [§4.3, Remark 4.3] Remark 4.3's disclosure that constants are not uniform as k → 2 is welcome; consider adding one line noting whether c_k, C_k could in principle be tracked explicitly from the proofs, to set reader expectations.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: constructive non-adaptive upper bound built from external localization plus self-contained decoder-side refinement identities.

full rationale

Theorem 2.1 is an existence proof of a fully non-adaptive public-coin protocol whose sample complexity matches the known rate r_k. The derivation chain is: (i) import a non-adaptive O(σ)-localization block from Lau–Scarlett (Proposition 2.2 / their Theorem 16); (ii) condition on success so |c−μ|≲σ and E|X−c|^k≤τ^k; (iii) prove unbiased decoder-side identities (correlated thresholds, safe-phase telescope or continuous kernel reproduction) and crossing-supported second-moment bounds; (iv) allocate scales by the square-root envelope of those bounds; (v) concentrate via median-of-means. None of these steps defines the target rate in terms of itself, fits a free parameter to data and relabels it a prediction, or rests on a self-citation uniqueness theorem. Optimality in the small-error regime is matched to Lau–Scarlett’s external lower bound (their Theorem 9), not forced by renaming. Numerical Appendix C only validates second-moment envelopes; it does not enter the proof. Dependency on an external localization theorem is ordinary citation, not circularity.

Assumptions & free parameters 0 free parameters · 6 assumptions · 2 invented entities

The result is a constructive information-theoretic upper bound. It imports standard concentration and one external localization theorem, assumes the nonparametric moment class D(k,λ,σ), and works in the public-coin arbitrary-measurable-query model. No data-fitted constants enter the rate; k-dependent absolute constants are existence constants from the proofs.

assumptions (6)
  • domain assumption Non-adaptive localization: O(ℓ(λ/σ)+log(1/η)) fixed 1-bit queries return an O(σ)-length interval containing the mean w.p. ≥1−η under a first-moment bound (Lau–Scarlett Theorem 16 / Prop. 2.2).
    Imported black-box; refinement analysis conditions on its success and never re-proves it.
  • domain assumption Matching Ω_k(r_k) lower bound for arbitrary (even adaptive) 1-bit protocols in a stated small-error, high-confidence range (Lau–Scarlett Theorem 9).
    Used only for minimax optimality corollary, not for the upper-bound construction.
  • domain assumption Class D(k,λ,σ): |EX|≤λ and E|X−EX|^k ≤σ^k for fixed k>1; samples i.i.d.; public coins independent of data.
    Problem setup Section 1.1; Lyapunov gives the first-moment bound needed by localization.
  • standard math Median-of-means / Chebyshev block concentration for bounded second-moment random variables (Lemma A.1).
    Standard; converts second-moment bounds into (ε,δ) sample counts.
  • domain assumption Query model allows arbitrary Borel subsets of R fixed by public coins before any message; decoder may use full transcript (Definition 1.1).
    Essential: threshold/interval restrictions retain an interaction gap per prior work.
  • standard math Correlated threshold / public-threshold identity: R(B−B_c) unbiased for h(X)−h(c) with second moment R|h(X)−h(c)| (Lemma 3.2; related to Mayekar et al. Wyner–Ziv estimators).
    Elementary uniform-threshold calculation; used as the scalar decoder-side recentering primitive.
invented entities (2)
  • Safe periodic residues with adjacent-scale telescope (r_j, f_j) and decoder-selected phases b_L(c) independent evidence
    purpose: Allow one fixed batch of periodic queries to serve every possible coarse center while confining aliasing to a terminal tail and making scale-L variance crossing-supported.
    Core algebraic device of the dyadic construction (Section 3.1–3.4, Lemma 4.1). Not a physical entity; a proof technique with explicit Borel implementation in Appendix D.
  • Continuous-scale random-grid refinement with Rademacher cell coloring and compact kernel ψ_a independent evidence
    purpose: Second, non-dyadic reconstruction identity that integrates a compactly supported kernel over scales to reproduce X−c at the decoder.
    Section 5 / Appendix B; complementary construction reaching the same rates. Falsifiable via the stated second-moment bounds and numerics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Universal Refinement without Interaction: Order-Optimal 1-Bit Mean Estimation." pith.science (2026). https://pith.science/paper/MICIARZ7

@misc{pith2026260724358,
  author       = {Pith},
  title        = {Pith review of: Universal Refinement without Interaction: Order-Optimal 1-Bit Mean Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MICIARZ7}},
  note         = {Machine review of arXiv:2607.24358}
}
abstract

This paper shows that interaction is unnecessary for order-optimal 1-bit mean estimation under finite central moments. For distributions satisfying $|\mathbb{E}X|\leq\lambda$ and $\mathbb{E}|X-\mathbb{E}X|^k\leq\sigma^k$ for a fixed $k>1$, we construct a fully non-adaptive public-coin protocol that fixes every measurable 1-bit query before communication. All localization and refinement queries are generated in a single batch; a subsequently decoded coarse center changes only how the stored refinement bits are interpreted. Two complementary constructions realize this decoder-side refinement: a finite dyadic scheme based on periodic residues and a continuous-scale scheme based on shifted random grids. Up to $k$-dependent constants, the refinement cost is $(\sigma/\epsilon)^2\log(1/\delta)$ for $k>2$, $(\sigma/\epsilon)^2[1+\log(\sigma/\epsilon)]\log(1/\delta)$ for $k=2$, and $(\sigma/\epsilon)^{k/(k-1)}\log(1/\delta)$ for $1<k<2$. Together with the additive localization cost $1+\log(\lambda/\sigma)$, these rates answer the Lau--Scarlett open problem for arbitrary measurable 1-bit queries in the affirmative. In the parameter range covered by existing small-error, high-confidence lower bounds, the resulting sample complexity is minimax optimal.

Figures

Figures reproduced from arXiv: 2607.24358 by the authors.

Figure 1
Figure 1. Safe phase at one scale. The boundary grids GL,0 and GL,1 differ by half a period. At least one phase has no boundary in the L/4-neighborhood of c (gray), so its residue is linear throughout that neighborhood. joining c and any x with |x − c| < L/4 contains no boundary, so the residue has slope one throughout that interval [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Continuous-scale refinement. (a) With c in the central half of cell Q, color averaging gives responses −1, 0, and +1 on the left adjacent, central, and right adjacent cells, respectively, and zero on all other cells. (b) Shift averaging yields the compact kernel ψ1/4(|d|/r), whose normalized all-scale integral in (31) equals d = X − c. Moreover, conditional on its public seed, every encoder bit is the indicator of a… view at source ↗
Figure 3
Figure 3. Rao–Blackwellized numerical evaluation of the refinement rate. Each point uses 400,000 independent data draws; vertical bars are 95% normal intervals computed from the conditional second-moment integrand. The ordinate is (EW2 0 + EW2 )/(ϵ 2 vk) for the dyadic construction and EZ 2 c /(ϵ 2 vk) for the continuous construction. where each rj uses the safe phase selected by c. Lemma 3.2 and the phase probabilities give … view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Deterministic verification of the scale allocation at σ = 1, µ = 0.2, c = 0, and ϵ = 0.1. Bars show V(p)/V(p ⋆ ) from (65); p ⋆ is the matched allocation. Hatching and gray levels distinguish candidate laws without relying on color. The logarithmic ordinate exposes the…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Interaction Is Not Necessary for Order-Optimal 1-Bit Mean Estimation

    stat.ML 2026-08 conditional novelty 7.0 of 10

    A fully non-adaptive one-bit protocol — every query fixed before data arrives — matches the minimax-optimal adaptive sample complexity for mean estimation under finite k-th moments, answering the COLT 2026 open proble...

Reference graph

Works this paper leans on

31 extracted references · 3 canonical work pages · cited by 1 Pith paper

  1. [1]

    Robust mean estimation under quantization

    Pedro Abdalla and Junren Chen. Robust mean estimation under quantization. arXiv preprint arXiv:2601.07074, 2026

    arXiv 2026

  2. [2]

    Canonne, Yuhan Liu, Ziteng Sun, and Himanshu Tyagi

    Jayadev Acharya, Cl \'e ment L. Canonne, Yuhan Liu, Ziteng Sun, and Himanshu Tyagi. Interactive inference under information constraints. IEEE Transactions on Information Theory, 68 0 (1): 0 502--516, 2022 a . doi:10.1109/TIT.2021.3123905

    arXiv 2022

  3. [3]

    Canonne, Ziteng Sun, and Himanshu Tyagi

    Jayadev Acharya, Cl \'e ment L. Canonne, Ziteng Sun, and Himanshu Tyagi. The role of interactivity in structured estimation. In Proceedings of the Thirty-Fifth Conference on Learning Theory, volume 178 of Proceedings of Machine Learning Research, pages 1328--1355. PMLR, 2022 b

    2022

  4. [4]

    Canonne, Ziteng Sun, and Himanshu Tyagi

    Jayadev Acharya, Cl \'e ment L. Canonne, Ziteng Sun, and Himanshu Tyagi. Unified lower bounds for interactive high-dimensional estimation under information constraints. In Advances in Neural Information Processing Systems, volume 36, pages 51133--51165. Curran Associates, Inc., 2023. doi:10.52202/075280-2226

    work page doi:10.52202/075280-2226 2023

  5. [5]

    Lower bounds for learning distributions under communication constraints via F isher information

    Leighton Pate Barnes, Yanjun Han, and Ayfer \"O zg \"u r. Lower bounds for learning distributions under communication constraints via F isher information. Journal of Machine Learning Research, 21 0 (236): 0 1--30, 2020

    2020

  6. [6]

    Boufounos

    Petros T. Boufounos. Universal rate-efficient scalar quantization. IEEE Transactions on Information Theory, 58 0 (3): 0 1861--1872, 2012. doi:10.1109/TIT.2011.2173899

    arXiv 2012

  7. [7]

    Nguyen, and David P

    Mark Braverman, Ankit Garg, Tengyu Ma, Huy L. Nguyen, and David P. Woodruff. Communication lower bounds for statistical estimation problems via a distributed data processing inequality. In Proceedings of the 48th Annual ACM Symposium on Theory of Computing, pages 1011--1020. ACM, 2016

    2016

  8. [8]

    Tony Cai and Hongji Wei

    T. Tony Cai and Hongji Wei. Distributed gaussian mean estimation under communication constraints: Optimal rates and communication-efficient algorithms. Journal of Machine Learning Research, 25 0 (37): 0 1--63, 2024

    2024

Show all 31 references
  1. [9]

    Optimal mean estimation without a variance

    Yeshwanth Cherapanamjeri, Nilesh Tripuraneni, Peter Bartlett, and Michael Jordan. Optimal mean estimation without a variance. In Proceedings of the Thirty-Fifth Conference on Learning Theory, volume 178 of Proceedings of Machine Learning Research, pages 356--357. PMLR, 2022

    2022

  2. [10]

    Interaction is necessary for distributed learning with privacy or communication constraints

    Yuval Dagan and Vitaly Feldman. Interaction is necessary for distributed learning with privacy or communication constraints. In Proceedings of the 52nd Annual ACM SIGACT Symposium on Theory of Computing, pages 450--462. ACM, 2020. doi:10.1145/3357713.3384315

    2020

  3. [11]

    Optimality in mean estimation: Beyond worst-case, beyond sub- G aussian, and beyond 1+ moments

    Trung Dang, Jasper Lee, Maoyuan Raymond Song, and Paul Valiant. Optimality in mean estimation: Beyond worst-case, beyond sub- G aussian, and beyond 1+ moments. In Advances in Neural Information Processing Systems, volume 36, pages 4150--4176. Curran Associates, Inc., 2023. doi...

    2023 doi

  4. [12]

    Oliveira

    Luc Devroye, Matthieu Lerasle, Gabor Lugosi, and Roberto I. Oliveira. Sub- G aussian mean estimators. The Annals of Statistics, 44 0 (6): 0 2695--2725, 2016. doi:10.1214/16-AOS1440

    2016 doi

  5. [13]

    Lower bounds for locally private estimation via communication complexity

    John Duchi and Ryan Rogers. Lower bounds for locally private estimation via communication complexity. In Proceedings of the Thirty-Second Conference on Learning Theory, volume 99 of Proceedings of Machine Learning Research, pages 1161--1191. PMLR, 2019

    2019

  6. [14]

    Dealing with range anxiety in mean estimation via statistical queries

    Vitaly Feldman. Dealing with range anxiety in mean estimation via statistical queries. In Proceedings of the 28th International Conference on Algorithmic Learning Theory, volume 76 of Proceedings of Machine Learning Research, pages 629--640. PMLR, 2017

    2017

  7. [15]

    Locally private hypothesis selection

    Sivakanth Gopi, Gautam Kamath, Janardhan Kulkarni, Aleksandar Nikolov, Zhiwei Steven Wu, and Huanyu Zhang. Locally private hypothesis selection. In Proceedings of the Thirty-Third Conference on Learning Theory, volume 125 of Proceedings of Machine Learning Research, pages 1785...

    2020

  8. [16]

    Geometric lower bounds for distributed parameter estimation under communication constraints

    Yanjun Han, Ayfer \"O zg \"u r, and Tsachy Weissman. Geometric lower bounds for distributed parameter estimation under communication constraints. In Proceedings of the 31st Conference on Learning Theory, volume 75 of Proceedings of Machine Learning Research, pages 3163--3188. ...

    2018

  9. [17]

    The sample complexity of distributed simple binary hypothesis testing under information constraints

    Hadi Kazemi, Ankit Pensia, and Varun Jog. The sample complexity of distributed simple binary hypothesis testing under information constraints. In Proceedings of the Thirty-Eighth Conference on Learning Theory, volume 291 of Proceedings of Machine Learning Research, pages 3213-...

    2025

  10. [18]

    Alon Kipnis and John C. Duchi. Mean estimation from one-bit measurements. IEEE Transactions on Information Theory, 68 0 (9): 0 6276--6296, 2022. doi:10.1109/TIT.2022.3175608

    2022

  11. [19]

    One-bit distributed mean estimation with unknown variance

    Ritesh Kumar and Shashank Vatedka. One-bit distributed mean estimation with unknown variance. Transactions on Machine Learning Research, 2026

    2026

  12. [20]

    Ivan Lau and Jonathan Scarlett. Open problem: Is interaction necessary for order-optimal 1-bit mean estimation? In Proceedings of the Thirty-Ninth Conference on Learning Theory, volume 336 of Proceedings of Machine Learning Research, pages 7123--7128. PMLR, 2026 a

    2026

  13. [21]

    Order-optimal sequential 1-bit mean estimation in general tail regimes

    Ivan Lau and Jonathan Scarlett. Order-optimal sequential 1-bit mean estimation in general tail regimes. arXiv preprint arXiv:2604.07796, 2026 b . Version 2, revised 22 May 2026

    2026 arXiv

  14. [22]

    Sequential 1-bit mean estimation with near-optimal sample complexity

    Ivan Lau and Jonathan Scarlett. Sequential 1-bit mean estimation with near-optimal sample complexity. In Proceedings of the 29th International Conference on Artificial Intelligence and Statistics, volume 300 of Proceedings of Machine Learning Research. PMLR, 2026 c

    2026

  15. [23]

    Jasper C. H. Lee and Paul Valiant. Optimal sub- G aussian mean estimation in R . In IEEE Annual Symposium on Foundations of Computer Science, pages 672--683. IEEE, 2022

    2022

  16. [24]

    Universal decentralized estimation in a bandwidth constrained sensor network

    Zhi-Quan Luo. Universal decentralized estimation in a bandwidth constrained sensor network. IEEE Transactions on Information Theory, 51 0 (6): 0 2210--2219, 2005

    2005

  17. [25]

    Wyner-- Z iv estimators: Efficient distributed mean estimation with side-information

    Prathamesh Mayekar, Ananda Theertha Suresh, and Himanshu Tyagi. Wyner-- Z iv estimators: Efficient distributed mean estimation with side-information. In Proceedings of the 24th International Conference on Artificial Intelligence and Statistics, volume 130 of Proceedings of Mac...

    2021

  18. [26]

    Efficient median of means estimator

    Stanislav Minsker. Efficient median of means estimator. In Proceedings of the Thirty-Sixth Conference on Learning Theory, volume 195 of Proceedings of Machine Learning Research, pages 5925--5933. PMLR, 2023

    2023

  19. [27]

    Pour, Hassan Ashtiani, and Shahab Asoodeh

    Alireza F. Pour, Hassan Ashtiani, and Shahab Asoodeh. Sample-optimal locally private hypothesis selection and the provable benefits of interactivity. In Proceedings of the Thirty-Seventh Conference on Learning Theory, volume 247 of Proceedings of Machine Learning Research, pag...

    2024

  20. [28]

    Giannakis

    Alejandro Ribeiro and Georgios B. Giannakis. Bandwidth-constrained distributed estimation for wireless sensor networks--- Part I : G aussian case. IEEE Transactions on Signal Processing, 54 0 (3): 0 1131--1143, 2006 a

    2006

  21. [29]

    Giannakis

    Alejandro Ribeiro and Georgios B. Giannakis. Bandwidth-constrained distributed estimation for wireless sensor networks--- Part II : Unknown probability density function. IEEE Transactions on Signal Processing, 54 0 (7): 0 2784--2796, 2006 b

    2006

  22. [30]

    Fundamental limits of online and distributed algorithms for statistical learning and estimation

    Ohad Shamir. Fundamental limits of online and distributed algorithms for statistical learning and estimation. In Advances in Neural Information Processing Systems, volume 27. Curran Associates, Inc., 2014

    2014

  23. [31]

    Information-theoretic lower bounds for distributed statistical estimation with communication constraints

    Yuchen Zhang, John Duchi, Michael Jordan, and Martin J Wainwright. Information-theoretic lower bounds for distributed statistical estimation with communication constraints. In C.J. Burges, L. Bottou, M. Welling, Z. Ghahramani, and K. Weinberger, editors, Advances in Neural Inf...

    2013

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.