Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Minimum-norm interpolation achieves benign overfitting for ℓp norms under sub-Gaussian covariates, provided the norm is 2-uniformly convex.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 17:04 UTC pith:JDTTRJWK

load-bearing objection Serious geometric framework, but the headline sub-Gaussian ℓ_p bounds rest on Assumption 2, which is never actually proved — the non-Gaussian extension is conditional, not established. the 4 major comments →

arxiv 2603.28956 v2 pith:JDTTRJWK submitted 2026-03-30 math.FA cs.LGmath.MGmath.PRmath.STstat.TH

Minimum Norm Interpolation via the Local Theory of Banach Spaces: The Role of 2-Uniform Convexity

Gil Kur , Pierre Bizeul This is my paper
classification math.FA cs.LGmath.MGmath.PRmath.STstat.TH MSC 46B2046B09
keywords minimum norm interpolationbenign overfitting2-uniform convexityℓp regressionsub-Gaussian covariatesK-convexityBanach space geometryoverparameterized linear regression
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper aims to show that the minimum-norm interpolator (MNI) generalizes well in overparameterized regression whenever the norm of the hypothesis space is 2-uniformly convex, a curvature condition far weaker than being induced by an inner product. It proves an upper bound on the structural (bias) error under this condition, and shows the bound is sharp for linear models when the unit ball sits in isotropic or John's position and covariates are i.i.d. symmetric sub-Gaussian. The main payoff is for ℓp-MNI with p just above 1: the noise error is eΘ(n/d), matching the Gaussian-covariate rate and establishing benign overfitting for a non-Hilbert norm without Gaussian assumptions. A sympathetic reader should care because this is the first sharp analysis of non-inner-product MNIs beyond Gaussian covariates, and it ties generalization to classical Banach-space geometry.

Core claim

On the paper's own terms, the central discovery is that 2-uniform convexity of the norm controls the bias of the MNI: the structural error T1 is O(R_{MM*} G_n(F)/t), where t is the uniform-convexity constant and R_{MM*} is the mean-width ratio of the projected unit ball. In linear models with the unit ball in isotropic or John's position and i.i.d. symmetric sub-Gaussian covariates, this bound becomes sharp, T1 = O~((G_n(K)^2 + 1/n)/t^2). For the ℓp-MNI with p near 1 and O(1)-sparse ground truth, the noise error satisfies T2 = eΘ(n/d), the same rate as the Gaussian case, and the paper proves this without any inner-product structure or closed-form solution.

What carries the argument

The load-bearing object is the 2-uniform convexity of the norm, a quantitative parallelogram-type inequality with constant t = p−1 for ℓp when p ∈ (1,2], which replaces the inner product as the source of curvature. Around it, the argument is carried by the M M* ratio R_{MM*} = M_n(F) M*_n(F)/n, the product of the mean norms of the projected unit ball and its polar, together with the K-convexity constant, which together control the bias-variance decomposition T1+T2. A variational interpolation argument plus a finite-volume-ratio covering lemma for sub-Gaussian projections supplies the sharp ℓp rates.

Load-bearing premise

Everything rests on Assumption 2: with high probability the norm of the interpolator of pure noise is bounded below by a universal constant; this inductive-bias condition is verified only for ℓp in a restricted dimension range, and if pure noise can be fit with small norm, the bias-variance bounds collapse.

What would settle it

Simulate the ℓp-MNI with p = 1 + C/log log d, i.i.d. Rademacher covariates, d ≍ n exp(C p(d)^{-2}), and zero ground truth; compute E||ŵ||_2^2 over many noise draws. If the noise error deviates from Θ(n/d) by more than a polylogarithmic factor, or if ∥ŵ(X,ξ)∥ falls below the constant C with probability larger than n^{-2}, then Theorem 4 or Assumption 2 fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If Theorem 4 is correct, the ℓp-MNI with p just above 1 achieves noise error eΘ(n/d) under i.i.d. Rademacher or other symmetric sub-Gaussian covariates, matching the rate previously known only for Gaussian data.
  • Theorem 2 implies the structural error of the MNI decays at the fast 1/n rate, up to logarithmic factors, for any 2-uniformly convex norm in isotropic or John's position, not just Hilbert norms.
  • Theorem 3 gives a general lower bound on the noise-driven variance of the MNI in terms of the K-convexity constant, so the n/d rate is intrinsic to the interpolation problem, not an artifact of ℓp geometry.
  • Corollary 1 shows the bias term T1 = eΘ(d^{2p−2}/n^p), identifying exactly where the exponent p enters the mean squared error.

Where Pith is reading between the lines

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

  • The same geometric mechanism suggests that the noise-error rate should hold for any 2-uniformly convex norm placed in the right position, provided sharp covering estimates for that position exist, a step the authors explicitly leave open.
  • Because the proof of Theorem 4 only needs curvature on average, one can conjecture that cotype 2, together with a tailored position, is the true minimal assumption; the paper itself floats this conjecture.
  • The techniques could transfer to nonparametric settings such as Sobolev-space MNIs, where a suitable norm combination preserves 2-uniform convexity and may yield minimax-optimal interpolators.
  • A testable consequence for practitioners: ℓp-MNI with p slightly below 2 should be roughly as harmless as ridge interpolation on sub-Gaussian features, without requiring Gaussian design matrices.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper studies minimum-norm interpolation (MNI) in 2-uniformly convex Banach spaces. It proposes a general upper bound on the structural error T1 under UC(2) (Theorem 1), a sharp localized bound in linear models when the unit ball is in isotropic or John's position and covariates are sub-Gaussian (Theorem 2), and a lower bound on the noise term via K-convexity and cotype-2 (Theorem 3). For the ℓ_p-MNI specifically, with p close to 2 and d growing faster than n, it claims sharp bounds T2 = eΘ(n/d) under i.i.d. symmetric sub-Gaussian covariates (Theorem 4 and Corollary 1), extending earlier Gaussian-covariate results of Donhauser et al. The proof combines Dvoretzky/Kashin-type inclusions, entropy estimates, convex concentration, and a variational argument over dyadic blocks. The manuscript is an extended version of the authors' ICML 2024 preliminary work.

Significance. If the main claims are fully established, this would be the first sharp benign-overfitting result for MNI under a non-Hilbert norm with non-Gaussian covariates, matching the known Gaussian rate n/d. The geometric approach — using UC(2), M·M* estimates, K-convexity, and positions of convex bodies — is promising and is a genuine conceptual contribution. The paper also usefully identifies the role of inductive-bias assumptions and positions. However, the advertised non-Gaussian extension currently rests on an unverified inductive-bias assumption (Assumption 2), and there is a mismatch between Corollary 1 and its proof. The significance is therefore conditional on closing these gaps.

major comments (4)
  1. [§1.1, Eq. (5); §7.1.5] Assumption 2 is load-bearing for Theorem 4, but in the sub-Gaussian part of the proof the required lower bound on ∥bfn(X,ξ)∥ = ∥ξ∥_n is never established. Step V (Section 7.1.5) obtains the inclusion cB^n ⊂ P eB^d_p (Eq. (43)) and an upper bound on M*(P eB^d_p) via Lemma 7; the lower bound on ∥ξ∥_n is only an indirect consequence of M·M* ≥ 1, which yields a bound of order α_d^{-1}, and no argument shows this is ≥ C in the theorem's regime. Since Theorem 4's conclusion depends on Assumption 2, the theorem as stated is not proved. Please either supply a direct verification of Assumption 2 for the sub-Gaussian case, or state Theorem 4 with Assumption 2 as an explicit condition and adjust the conclusion accordingly.
  2. [§3.1, Corollary 1; §7.3.3] Corollary 1 states T1 = eΘ(d^{2p-2}/n^p), but its proof in §7.3.3 only establishes the upper bound T1 ≲ Õ(Gn(B_p^d)^{2p}) = Õ(d^{2p-2}/n^p). No matching lower bound is derived, so the eΘ claim is unsupported. The statement should be weakened to an O/Õ upper bound, or the missing lower bound should be proved.
  3. [§7.1.1, after Eq. (26), footnote 7] The uniform estimate (26) over the sets V_k, which are not finite, is asserted while the footnote says 'the statement requires an entirely elementary net argument which is omitted.' This estimate is used to control all dyadic blocks in Step II and thus supports the main variational argument of Theorem 4. The omitted net argument is load-bearing and should be supplied before the result can be considered proved.
  4. [§3, Lemma 1 and Remark 3; §4.3] Several results depend on works that are either not yet published or coauthored by the present authors: Lemma 1 and Remark 3 rely on Bizeul (2025) and Bizeul–Klartag (2025) for the slicing/isotropic-position estimates, Remark 2 defers to Kur et al. (2026), and §4.3 invokes a quantitative Anderson inequality from Bizeul et al. (2026, 'to appear'). These dependencies are not part of the manuscript and are not independently verifiable from the text. For a journal submission, either include the necessary statements and proofs, or clearly mark them as external assumptions. This particularly affects the completeness of Theorem 2.
minor comments (5)
  1. [§6, Notation] The notation α_d := sqrt(p/(p-1)) eα_d := α_d^{-1} is confusing; α_d and eα_d should be defined separately and unambiguously.
  2. [§7.3, Lemma 1 proof] There are spelling inconsistencies: 'Blashke–Santalo' and 'Blackhe-Santalo' both appear. Please standardize to 'Blaschke–Santalo'.
  3. [§3.1 and §6] The parameter p(d) is used both as C/log log d and, in Theorem 4, via the expression 1 + C log log log(d)·p(d). This is notationally confusing and should be clarified.
  4. [§7.4.1] In the proof of Theorem 3, the sentence 'note that for each realization X and ξ, it holds fLn|X = ξ' appears to contain a typo; the linearized operator should be defined and evaluated precisely.
  5. [References] Several references are marked 'to appear' or are not clearly peer-reviewed (Bizeul et al. 2026, Kur et al. 2026, Karhadkar et al. 2026). The reader would benefit from availability information or versions at the time of submission.

Circularity Check

0 steps flagged

No definitional circularity: the main bound is derived from UC(2)/K-convexity geometry; the fragile point is an unverified Assumption 2, not a self-referential reduction.

full rationale

The central derivation is not circular. Theorem 1's bias bound follows from UC(2) quadrature plus Gaussian norm concentration; Theorem 2's T1 bound follows from Lemma 1 and a variational comparison; Theorem 4's T2 upper bound comes from a coordinate-block variational argument (Eqs. 35–39), and its lower bound from the reverse Efron-Stein/K-convexity estimate (Theorem 3). None of these equations is shown to equal an input by construction, and no parameter is fitted to the claimed rate. The main genuine weakness is Assumption 2 (Eq. 5): Theorem 4's eTheta(n/d) claim and Theorem 3 both use it, but the sub-Gaussian proof only supplies the inclusion cB^n ⊂ P eB^d_p and M*(P eB^d_p) ≲ α_d, giving a lower bound of order 1/α_d, which tends to 0 in the stated p→1 regime. The text itself notes 'the upper and lower bounds on n are essential for Assumption 2 to hold' but never verifies that assumption in that regime. This is a missing-verification/correctness gap, not a circular reduction: the assumption is not derived from the conclusion, and the theorem is conditional on it. Minor self-references appear—Remark 2 defers claims to the first author's forthcoming Kur et al. (2026), Section 4.3 cites Bizeul et al. (2026), and Lemma 1's proof cites Bizeul (2025) alongside the independent, published Klartag–Lehec (2025) resolution—but these are not load-bearing for the main theorem's arithmetic and do not turn the derivation into a tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 10 axioms · 0 invented entities

The paper introduces no fitted parameters and no new physical/mathematical entities. Its results rest on explicit domain assumptions about the norm, the signal, the covariates, and the dimension regime, plus several external theorems from convex geometry and Banach space theory.

axioms (10)
  • domain assumption 2-uniform convexity of (B(X),∥·∥) with constant t>0 (Definition UC(2), §1.2)
    Core curvature hypothesis for Theorems 1 and 2; weaker than Hilbert but stronger than cotype 2.
  • domain assumption Assumption 1: ∥f*∥_{L2(P)} ≍ ∥f*∥ ≍ 1
    Rules out vanishing signal norm; stated in Section 1.1.
  • domain assumption Assumption 2: with probability ≥1−n^{-2}, ∥bfn(X,ξ)∥ ≥ C (§1.1, Eq. 5)
    Inductive-bias condition that the norm of the MNI on pure noise is not too small; load-bearing for all theorems.
  • domain assumption Assumption 3: Mg(Fn) ≍ Mn(F) and M*g(Fn) ≍ M*n(F) with high probability (§2.2)
    Concentration of Gaussian means of the random projection of the unit ball.
  • domain assumption Assumption 4: lower isometry remainder IL(n,F,P) ≤ C4 G(F)^2 (§2.2)
    Allows passage from L1 to L2 control; needed in Theorem 1 and for small-ball arguments.
  • domain assumption Assumption 5: affine sections of K in direction w* are centrally symmetric (§3, before Theorem 2)
    Needed for the sharp localized bound; holds for ℓ_p-ball with 1-sparse w*.
  • domain assumption For Theorem 2: unit ball in isotropic or John's position; symmetric isotropic i.i.d. sub-Gaussian covariates; w* 1-sparse; d≥C1 n
    Position and covariate assumptions required for Lemma 1 and the sharp bound.
  • domain assumption For Theorem 4/Corollary 1: w* is O(1)-sparse; n log(d)^C ≤ d ≤ n^{q/2} log(d)^{-C}; sub-Gaussian covariates satisfy CCP with constant O(√log d)
    Dimension/sample range needed for Assumption 2 and for Kashin/CCP arguments to yield n/d rate.
  • standard math External theorem: affirmative resolution of Bourgain's slicing conjecture (Klartag-Lehec 2025; Bizeul 2025)
    Used in Lemma 1 and Remark 3 to place isotropic bodies in M-position and bound M*s(K). One of the resolving papers is by a coauthor.
  • standard math External theorem: Kashin's theorem for finite-volume-ratio bodies under sub-Gaussian projections (Litvak et al. 2004, 2005)
    Used in Lemma 3 and Step V of Theorem 5 to replace Dvoretzky's theorem for non-Gaussian covariates.

pith-pipeline@v1.3.0-alltime-deepseek · 43694 in / 15522 out tokens · 149331 ms · 2026-08-02T17:04:55.405790+00:00 · methodology

0 comments
read the original abstract

The minimum-norm interpolator (MNI) framework has recently attracted considerable attention as a tool for understanding generalization in overparameterized models, such as neural networks. In this work, we study the MNI under a $2$-uniform convexity assumption, which is weaker than requiring the norm to be induced by an inner product; in this setting, the MNI typically does not admit a closed-form solution. At a high level, we show that this condition yields an upper bound on the MNI bias in both linear and nonlinear models. We further show that this bound is sharp for overparameterized linear regression when the norm's unit ball is in isotropic or John's position and the covariates are i.i.d.\ sub-Gaussian, for example, when each covariate vector has i.i.d. Rademacher entries. Finally, under the same assumption on the covariates, we prove sharp generalization bounds for the $\ell_p$-MNI when $p \in \bigl(1 + C/\log d, 2\bigr]$. To the best of our knowledge, this is the \emph{first} work to establish sharp bounds for non-Gaussian covariates in linear models when the norm is not induced by an inner product. This work is deeply inspired by classical work on $K$-convexity and recent work on the geometry of $2$-uniformly convex and isotropic convex bodies.

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. Minimum Norm Interpolation via The Local Theory of Banach Spaces: The Role of Gaussianity

    math.ST 2026-07 conditional novelty 6.0

    The sharp MSE bound for the ℓ1-minimum-norm interpolator under isotropic Gaussian covariates is recovered via the geometry of symmetric Gaussian polytopes, without the convex Gaussian min-max theorem.

Reference graph

Works this paper leans on

14 extracted references · 6 linked inside Pith · cited by 1 Pith paper

  1. [7]

    Vladimir Koltchinskii.Oracle Inequalities in Empirical Risk Minimization and Sparse Recovery Problems: Ecole d’Eté de Probabilités de Saint-Flour XXXVIII-2008, volume

    2008

  2. [8]

    A new perspective on minimum norm interpolation under gaussian covariates.To appear at AISTATS 2026,

    Gil Kur, Paul Smanjuntak, Zong Shang, Guillaume Lecué, and Reese Pathak. A new perspective on minimum norm interpolation under gaussian covariates.To appear at AISTATS 2026,

    2026

  3. [10]

    Extending the scope of the small-ball method.arXiv preprint arXiv:1709.00843,

    Shahar Mendelson. Extending the scope of the small-ball method.arXiv preprint arXiv:1709.00843,

    Pith/arXiv arXiv

  4. [12]

    Deep double descent: Where bigger models and more data hurt.Journal of Statistical Mechanics: Theory and Experiment, 2021(12):124003,

    Preetum Nakkiran, Gal Kaplun, Yamini Bansal, Tristan Yang, Boaz Barak, and Ilya Sutskever. Deep double descent: Where bigger models and more data hurt.Journal of Statistical Mechanics: Theory and Experiment, 2021(12):124003,

    2021

  5. [1936]

    Fast rates for noisy interpolation require rethinking the effects of inductive bias.arXiv preprint arXiv:2203.03597,

    Konstantin Donhauser, Nicolo Ruggeri, Stefan Stojanovic, and Fanny Yang. Fast rates for noisy interpolation require rethinking the effects of inductive bias.arXiv preprint arXiv:2203.03597,

    Pith/arXiv arXiv

  6. [1977]

    On the duality between type and cotype

    Gilles Pisier. On the duality between type and cotype. InMartingale Theory in Harmonic Analysis and Banach Spaces: Proceedings of the NSF-CBMS Conference Held at the Cleveland State University, Cleveland, Ohio, July 13–17, 1981, pages 131–144. Springer,

    1981

  7. [1999]

    Probabilistic methods in the geometry of banach spaces

    Gilles Pisier. Probabilistic methods in the geometry of banach spaces. InProbability and Analysis: Lectures given at the 1st 1985 Session of the Centro Internazionale Matematico Estivo (CIME) held at Varenna (Como), Italy May 31–June 8, 1985, pages 167–241. Springer,

    1985

  8. [2009]

    On the mean-width of isotropic convex bodies and their associated l p-centroid bodies

    50 Emanuel Milman. On the mean-width of isotropic convex bodies and their associated l p-centroid bodies. International Mathematics Research Notices, 2015(11):3408–3423,

    2015

  9. [2014]

    On the robustness of minimum norm interpolators and regularized empirical risk minimizers.arXiv preprint arXiv:2012.00807,

    Geoffrey Chinot, Matthias Löffler, and Sara van de Geer. On the robustness of minimum norm interpolators and regularized empirical risk minimizers.arXiv preprint arXiv:2012.00807,

    Pith/arXiv arXiv 2012

  10. [2019]

    The slicing conjecture via small ball estimates.arXiv preprint arXiv:2501.06854,

    Pierre Bizeul. The slicing conjecture via small ball estimates.arXiv preprint arXiv:2501.06854,

    Pith/arXiv arXiv

  11. [2020]

    Harmful overfitting in sobolev spaces.arXiv preprint arXiv:2602.00825,

    Kedar Karhadkar, Alexander Sietsema, Deanna Needell, and Guido Montufar. Harmful overfitting in sobolev spaces.arXiv preprint arXiv:2602.00825,

    arXiv

  12. [2021]

    On milman’s inequality and random subspaces which escape through a mesh inRn

    Yehoram Gordon. On milman’s inequality and random subspaces which escape through a mesh inRn. In Geometric Aspects of Functional Analysis: Israel Seminar (GAFA) 1986–87, pages 84–106. Springer,

    1986

  13. [2025]

    Distances between non-symmetric convex bodies: optimal bounds up to polylog.arXiv preprint arXiv:2510.20511,

    Pierre Bizeul and Boaz Klartag. Distances between non-symmetric convex bodies: optimal bounds up to polylog.arXiv preprint arXiv:2510.20511,

    Pith/arXiv arXiv

  14. [2026]

    A geometrical viewpoint on the benign overfitting property of the minimum ℓ2-norm interpolant estimator.arXiv preprint arXiv:2203.05873,

    Guillaume Lecué and Zong Shang. A geometrical viewpoint on the benign overfitting property of the minimum ℓ2-norm interpolant estimator.arXiv preprint arXiv:2203.05873,

    Pith/arXiv arXiv