arXiv subjects

Gil Kur

Publications and source records attributed to Gil Kur.

At least 19 recordsLinked to original sources

arXiv2026

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

We study minimum-norm interpolation (MNI) in overparameterized linear regression with isotropic Gaussian covariates, in settings where the MNI has no closed-form formula. Whereas most prior work relied on Gaussian comparison tools such as the convex Gaussian min--max theorem (CGMT), our approach uses tools from high-dimensional geometry and probability. First, when the norm is in isotropic position, we obtain an ``offset'' bound that controls the amount by which the MNI shrinks the ground truth. Second, we show that the ``intrinsic'' variance of the $\ell_1$-MNI is at most $O(\tfrac{1}{n\log(d/n)^2})$, using a variant of Talagrand's $L_1$--$L_2$ inequality due to Cordero-Erausquin and Ledoux [2012], together with a classical result of Gluskin [1988]. We recover the sharp mean-squared error (MSE) bound for the $\ell_1$-MNI obtained by Wang et al. [2022], using the work of Fleury [2012] on the symmetric Gaussian polytope, which is defined via \[ P_{n,d} := \mathrm{conv}\{\pm X_i\}_{i=1}^{d} \text{ where } X_i \overset{\mathrm{i.i.d.}}{\sim} N(0,\mathrm{I}_{n \times n}), \] rather than CGMT. Our methods also imply improvements on previous results in high-dimensional geometry that may be of independent interest. First, we show that with overwhelming probability, the ratio between the isotropic constant of $P_{n,d}$ and that of the Euclidean ball in $\mathbb{R}^n$ is at most $1+O((\log(d/n))^{-2})$, improving a result of Klartag and Kozma [2009]. We also establish a refined weighted thin-shell estimate on $P_{n,d}$, and provide an elementary proof of the main theorem of Fleury [2012].

math.ST↗

arXiv2026

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

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.

math.FA↗

arXiv2025

Revisiting Knowledge Distillation: The Hidden Role of Dataset Size

The concept of knowledge distillation (KD) describes the training of a student model from a teacher model and is a widely adopted technique in deep learning. However, it is still not clear how and why distillation works. Previous studies focus on two central aspects of distillation: model size, and generalisation. In this work we study distillation in a third dimension: dataset size. We present a suite of experiments across a wide range of datasets, tasks and neural architectures, demonstrating that the effect of distillation is not only preserved but amplified in low-data regimes. We call this newly discovered property the data efficiency of distillation. Equipped with this new perspective, we test the predictive power of existing theories of KD as we vary the dataset size. Our results disprove the hypothesis that distillation can be understood as label smoothing, and provide further evidence in support of the dark knowledge hypothesis. Finally, we analyse the impact of modelling factors such as the objective, scale and relative number of samples on the observed phenomenon. Ultimately, this work reveals that the dataset size may be a fundamental but overlooked variable in the mechanisms underpinning distillation.

cs.LG↗

arXiv2025

Specialization after Generalization: Towards Understanding Test-Time Training in Foundation Models

Recent empirical studies have explored the idea of continuing to train a model at test-time for a given task, known as test-time training (TTT), and have found it to yield significant performance improvements. However, there is limited understanding of why and when TTT is effective. Earlier explanations mostly focused on the observation that TTT may help when applied to out-of-distribution adaptation or used with privileged data. However, the growing scale of foundation models with most test data being in-distribution questions these explanations. We instead posit that foundation models remain globally underparameterized, with TTT providing a mechanism for specialization after generalization, focusing capacity on concepts relevant to the test task. Specifically, under the linear representation hypothesis, we propose a model in which TTT achieves a substantially smaller in-distribution test error than global training. We empirically validate our model's key assumptions by training a sparse autoencoder on ImageNet, showing that semantically related data points are explained by only a few shared concepts. Finally, we perform scaling studies across image and language tasks that confirm the practical implications of our model, identifying the regimes where specialization is most effective.

cs.LG↗

arXiv2025

Sharp Risk Bounds for Early-Stopping in Gaussian Linear Regression

We study early-stopped mirror descent (ESMD) for high-dimensional Gaussian linear regression over arbitrary convex bodies and design matrices, where the task is to minimize the in-sample mean squared error. Our main result shows that some of the sharpest risk bounds for the least squares estimator (LSE), based on the local Gaussian width, extend to ESMD. We derive sufficient conditions on the potential, expressed via the Minkowski functional, under which our result holds. These conditions allow us to construct new potentials and analyze existing ones. Our results then yield general sufficient conditions for minimax optimality of ESMD, provide a systematic comparison with the LSE, and establish the tightest known risk bound in the $\ell_1$-constrained setting.

cs.LG↗

arXiv2025

On the Optimality of Random Partial Sphere Coverings in High Dimensions

Given $N$ geodesic caps on the unit sphere in $\mathbb{R}^d$, whose total normalized surface area is one, what is the maximal proportion of the sphere that their union can cover? In this work, we provide an asymptotically sharp upper bound for an antipodal partial covering of the sphere by $N=N(d)$ congruent caps in the regime $N(d)\to\infty$ and $\ln N(d)=o(\sqrt d)$, showing that the maximum proportion covered approaches $1 - e^{-1}$ as $d\to\infty$. We discuss the relation of this result to the optimality of random polytopes in high dimensions, the limitations of our technique via the Gaussian surface area bounds of K. Ball and F. Nazarov, and its applications in computer science theory.

math.MG↗

arXiv2024

Convergence rates for estimating multivariate scale mixtures of uniform densities

The Grenander estimator is a well-studied procedure for univariate nonparametric density estimation. It is usually defined as the Maximum Likelihood Estimator (MLE) over the class of all non-increasing densities on the positive real line. It can also be seen as the MLE over the class of all scale mixtures of uniform densities. Using the latter viewpoint, Pavlides and Wellner~\cite{pavlides2012nonparametric} proposed a multivariate extension of the Grenander estimator as the nonparametric MLE over the class of all multivariate scale mixtures of uniform densities. We prove that this multivariate estimator achieves the univariate cube root rate of convergence with only a logarithmic multiplicative factor that depends on the dimension. The usual curse of dimensionality is therefore avoided to some extent for this multivariate estimator. This result positively resolves a conjecture of Pavlides and Wellner~\cite{pavlides2012nonparametric} under an additional lower bound assumption. Our proof proceeds via a general accuracy result for the Hellinger accuracy of MLEs over convex classes of densities. We also provide algorithms for computing the estimator, and illustrate performance on real and simulated datasets.

math.ST↗

arXiv2024

Debiased LASSO under Poisson-Gauss Model

Quantifying uncertainty in high-dimensional sparse linear regression is a fundamental task in statistics that arises in various applications. One of the most successful methods for quantifying uncertainty is the debiased LASSO, which has a solid theoretical foundation but is restricted to settings where the noise is purely additive. Motivated by real-world applications, we study the so-called Poisson inverse problem with additive Gaussian noise and propose a debiased LASSO algorithm that only requires $n \gg s\log^2p$ samples, which is optimal up to a logarithmic factor.

math.ST↗

arXiv2023

On the Variance, Admissibility, and Stability of Empirical Risk Minimization

It is well known that Empirical Risk Minimization (ERM) may attain minimax suboptimal rates in terms of the mean squared error (Birg\'e and Massart, 1993). In this paper, we prove that, under relatively mild assumptions, the suboptimality of ERM must be due to its large bias. Namely, the variance error term of ERM is bounded by the minimax rate. In the fixed design setting, we provide an elementary proof of this result using the probabilistic method. Then, we extend our proof to the random design setting for various models. In addition, we provide a simple proof of Chatterjee's admissibility theorem (Chatterjee, 2014, Theorem 1.4), which states that in the fixed design setting, ERM cannot be ruled out as an optimal method, and then we extend this result to the random design setting. We also show that our estimates imply the stability of ERM, complementing the main result of Caponnetto and Rakhlin (2006) for non-Donsker classes. Finally, we highlight the somewhat irregular nature of the loss landscape of ERM in the non-Donsker regime, by showing that functions can be close to ERM, in terms of $L_2$ distance, while still being far from almost-minimizers of the empirical loss.

math.ST↗

arXiv2022

Tyler's and Maronna's M-estimators: Non-Asymptotic Concentration Results

Tyler's and Maronna's M-estimators, as well as their regularized variants, are popular robust methods to estimate the scatter or covariance matrix of a multivariate distribution. In this work, we study the non-asymptotic behavior of these estimators, for data sampled from a distribution that satisfies one of the following properties: 1) independent sub-Gaussian entries, up to a linear transformation; 2) log-concave distributions; 3) distributions satisfying a convex concentration property. Our main contribution is the derivation of tight non-asymptotic concentration bounds of these M-estimators around a suitably scaled version of the data sample covariance matrix. Prior to our work, non-asymptotic bounds were derived only for Elliptical and Gaussian distributions. Our proof uses a variety of tools from non asymptotic random matrix theory and high dimensional geometry. Finally, we illustrate the utility of our results on two examples of practical interest: sparse covariance and sparse precision matrix estimation.

math.ST↗

arXiv2022

An Efficient Minimax Optimal Estimator For Multivariate Convex Regression

This work studies the computational aspects of multivariate convex regression in dimensions $d \ge 5$. Our results include the \emph{first} estimators that are minimax optimal (up to logarithmic factors) with polynomial runtime in the sample size for both $L$-Lipschitz convex regression, and $\Gamma$-bounded convex regression under polytopal support. Our analysis combines techniques from empirical process theory, stochastic geometry, and potential theory, and leverages recent algorithmic advances in mean estimation for random vectors and in distribution-free linear regression. These results provide the first efficient, minimax-optimal procedures for non-Donsker classes for which their corresponding least-squares estimator is provably minimax-suboptimal.

math.ST↗

arXiv2021

On the Minimal Error of Empirical Risk Minimization

We study the minimal error of the Empirical Risk Minimization (ERM) procedure in the task of regression, both in the random and the fixed design settings. Our sharp lower bounds shed light on the possibility (or impossibility) of adapting to simplicity of the model generating the data. In the fixed design setting, we show that the error is governed by the global complexity of the entire class. In contrast, in random design, ERM may only adapt to simpler models if the local neighborhoods around the regression function are nearly as complex as the class itself, a somewhat counter-intuitive conclusion. We provide sharp lower bounds for performance of ERM for both Donsker and non-Donsker classes. We also discuss our results through the lens of recent studies on interpolation in overparameterized models.

math.ST↗

arXiv2020

A bounded-noise mechanism for differential privacy

We present an asymptotically optimal $(\epsilon,\delta)$ differentially private mechanism for answering multiple, adaptively asked, $\Delta$-sensitive queries, settling the conjecture of Steinke and Ullman [2020]. Our algorithm has a significant advantage that it adds independent bounded noise to each query, thus providing an absolute error bound. Additionally, we apply our algorithm in adaptive data analysis, obtaining an improved guarantee for answering multiple queries regarding some underlying distribution using a finite sample. Numerical computations show that the bounded-noise mechanism outperforms the Gaussian mechanism in many standard settings.

cs.DS↗

arXiv2020

On Suboptimality of Least Squares with Application to Estimation of Convex Bodies

We develop a technique for establishing lower bounds on the sample complexity of Least Squares (or, Empirical Risk Minimization) for large classes of functions. As an application, we settle an open problem regarding optimality of Least Squares in estimating a convex set from noisy support function measurements in dimension $d\geq 6$. Specifically, we establish that Least Squares is mimimax sub-optimal, and achieves a rate of $\tilde{\Theta}_d(n^{-2/(d-1)})$ whereas the minimax rate is $\Theta_d(n^{-4/(d+3)})$.

math.ST↗

arXiv2020

Convex Regression in Multidimensions: Suboptimality of Least Squares Estimators

Under the usual nonparametric regression model with Gaussian errors, Least Squares Estimators (LSEs) over natural subclasses of convex functions are shown to be suboptimal for estimating a $d$-dimensional convex function in squared error loss when the dimension $d$ is 5 or larger. The specific function classes considered include: (i) bounded convex functions supported on a polytope (in random design), (ii) Lipschitz convex functions supported on any convex domain (in random design), (iii) convex functions supported on a polytope (in fixed design). For each of these classes, the risk of the LSE is proved to be of the order $n^{-2/d}$ (up to logarithmic factors) while the minimax risk is $n^{-4/(d+4)}$, when $d \ge 5$. In addition, the first rate of convergence results (worst case and adaptive) for the unrestricted convex LSE are established in fixed-design for polytopal domains for all $d \geq 1$. Some new metric entropy results for convex functions are also proved which are of independent interest.

math.ST↗

arXiv2019

Double descent in the condition number

In solving a system of $n$ linear equations in $d$ variables $Ax=b$, the condition number of the $n,d$ matrix $A$ measures how much errors in the data $b$ affect the solution $x$. Estimates of this type are important in many inverse problems. An example is machine learning where the key task is to estimate an underlying function from a set of measurements at random points in a high dimensional space and where low sensitivity to error in the data is a requirement for good predictive performance. Here we discuss the simple observation, which is known but surprisingly little quoted (see Theorem 4.2 in \cite{Brgisser:2013:CGN:2526261}): when the columns of $A$ are random vectors, the condition number of $A$ is highest if $d=n$, that is when the inverse of $A$ exists. An overdetermined system ($n>d$) as well as an underdetermined system ($n<d$), for which the pseudoinverse must be used instead of the inverse, typically have significantly better, that is lower, condition numbers. Thus the condition number of $A$ plotted as function of $d$ shows a double descent behavior with a peak at $d=n$.

cs.LG↗

arXiv2019

Intrinsic and Dual Volume Deviations of Convex Bodies and Polytopes

We establish estimates for the asymptotic best approximation of the Euclidean unit ball by polytopes under a notion of distance induced by the intrinsic volumes. We also introduce a notion of distance between convex bodies that is induced by the Wills functional, and apply it to derive asymptotically sharp bounds for approximating the ball in high dimensions. Remarkably, it turns out that there is a polytope which is almost optimal with respect to all intrinsic volumes simultaneously, up to absolute constants. Finally, we establish asymptotic formulas for the best approximation of smooth convex bodies by polytopes with respect to a distance induced by dual volumes, which originate from Lutwak's dual Brunn-Minkowski theory.

math.MG↗

arXiv2019

Optimality of Maximum Likelihood for Log-Concave Density Estimation and Bounded Convex Regression

In this paper, we study two problems: (1) estimation of a $d$-dimensional log-concave distribution and (2) bounded multivariate convex regression with random design with an underlying log-concave density or a compactly supported distribution with a continuous density. First, we show that for all $d \ge 4$ the maximum likelihood estimators of both problems achieve an optimal risk of $\Theta_d(n^{-2/(d+1)})$ (up to a logarithmic factor) in terms of squared Hellinger distance and $L_2$ squared distance, respectively. Previously, the optimality of both these estimators was known only for $d\le 3$. We also prove that the $\epsilon$-entropy numbers of the two aforementioned families are equal up to logarithmic factors. We complement these results by proving a sharp bound $\Theta_d(n^{-2/(d+4)})$ on the minimax rate (up to logarithmic factors) with respect to the total variation distance. Finally, we prove that estimating a log-concave density - even a uniform distribution on a convex set - up to a fixed accuracy requires the number of samples \emph{at least} exponential in the dimension. We do that by improving the dimensional constant in the best known lower bound for the minimax rate from $2^{-d}\cdot n^{-2/(d+1)}$ to $c\cdot n^{-2/(d+1)}$ (when $d\geq 2$).

math.ST↗