Arun K. Kuchibhotla

Arun Kumar Kuchibhotla

Evergreen Junior Career Professor · 2024–2027

I am an associate professor in the Department of Statistics & Data Science at Carnegie Mellon University, where I study how to quantify uncertainty when standard statistical assumptions are difficult to justify.

For example, researchers often use the same data to choose a regression model and estimate its coefficients. I develop confidence intervals that account for this choice. I also study how to obtain valid coverage when observations are dependent or estimators have nonstandard limiting distributions.

Arun Kumar Kuchibhotla in an alpine meadow with mountains in the background

Recent preprints

All research

On the Lévy concentration function of Gaussian quadratic forms with applications to second order U-statistics

Abhimanyu Choudhary, Arun Kumar Kuchibhotla

Abstract

We provide an upper-bound for the Lévy concentration function: Qₛ(ε) = supₓ∈ℝ P(x < S ≤ x + ε) where S is a weighted sum of noncentral chi-square random variables: S = ∑ₖ≥₁ [λₖ(Zₖ² − 1) + μₖZₖ] Here, {Zₖ} for k ≥ 1 is a sequence of independent standard Gaussian random variables and {λₖ} and {μₖ}, indexed by k ≥ 1, are real valued, square summable sequences. Random variables of this type often appear as limiting distributions of second order U-statistics. Our bound is adaptive, in that it recovers (up to constant factors) Gaussian type concentration function estimates if ‖λ‖₂ is negligible compared to ‖μ‖₂ and chi-square estimates if ‖μ‖₂ is negligible compared to ‖λ‖₂. Our bound generalizes existing bounds in various ways. In particular, we make no assumptions regarding the number of nonzero |λₖ| or the size of the minimal |λₖ|, nor do we make any assumptions on the signs of λₖ. Finally, we apply our bound to some examples of interest, specifically quadratic forms that arise in limit theorems for second-order U-statistics.

Author abstract on arXiv

Description

How much probability can a Gaussian quadratic form put in a short interval? The answer depends on the relative sizes of its linear and quadratic parts and helps analyze the limiting distributions of second-order U-statistics.

Berry-Esseen bounds for multivariate martingale difference sequences in the Kolmogorov distance

Weichen Wu, Dung Le, Arun Kumar Kuchibhotla, Alessandro Rinaldo

Abstract

We derive new Gaussian approximations for finite martingale difference sequences in ℝᵈ with respect to the Kolmogorov distance. Under appropriate conditions, our bounds exhibit a dependence of order n⁻¹ᐟ⁴ on the length of the sequence and of order polylog(d) on the dimension. As an application, we derive a high-dimensional Berry-Esseen bound over hyper-rectangles for martingale sequences generated from Markov chains.

Author abstract on arXiv

Description

If today’s increment has conditional mean zero given the past, sums can be approximately Gaussian even with dependence. The bounds quantify the approximation error for many coordinates at once and include a Markov-chain application.

On a Probability Inequality for Order Statistics with Applications to Bootstrap, Conformal Prediction, and more

Manit Paul, Arun Kumar Kuchibhotla

Abstract

“Behind every limit theorem, there is an inequality” said Kolmogorov. We say “for every inequality, there is an approximate inequality under approximate regularity conditions.” Suppose X, X' are independent and identically distributed random variables. Then X ≤ X′ with a probability of at least 1/2, irrespective of the underlying (common) distribution. One can ask what happens to the probability if X, X' are independent but not identically distributed. It should be approximately 1/2 if the distributions are approximately equal. Similarly, what if the random variables are dependent? It should, again, be approximately 1/2 if the random variables are approximately independent. We explore an extension of this probability inequality involving order statistics and develop approximate versions of such an inequality under violations of independence and identical distribution assumptions. We further show that this inequality can be used as a basis to prove asymptotic validity of bootstrap/subsampling, finite-sample validity of conformal prediction, permutation tests, and asymptotic validity of rank tests without group invariance. Specifically, in the context of resampling inference, our results can be seen as a finite-sample instantiation of some results by Peter Hall and yield an alternative “cheap bootstrap” that applies to high-dimensional data.

Author abstract on arXiv

Description

Bootstrap calibration, permutation tests, and conformal prediction all rely on how one observation ranks among others. This paper shows how departures from independence or identical distributions change the relevant rank probabilities.

Sharp Debiasing for Smooth Functional Estimation in Banach Spaces

Woonyoung Chang, Arun Kumar Kuchibhotla

Abstract

This paper studies the estimation of smooth functionals f(θ) of a mean parameter θ = 𝔼_P[W] for a distribution P on a general Banach space. We propose a cross-fitted estimator based on a single sample splitting and establish non-asymptotic moment bounds and Berry--Esséen bounds for both m-smooth and infinitely smooth functionals under the finite moment assumptions. Our framework is applied to precision matrix estimation and the inference of projection parameters in high-dimensional regression. In these Euclidean settings, the proposed estimators achieve asymptotic normality under the dimension regime d log²(en) = o(n) without requiring any structural assumptions (e.g., sparsity). We discuss computational relaxations that enables polynomial-time implementation for a range of matrix functionals.

Author abstract on arXiv

Description

Applying a smooth nonlinear function to a sample mean can introduce substantial bias in high dimensions. The paper corrects this bias and studies uncertainty for examples including precision matrices and regression projection parameters.

Time-uniform conformal and PAC prediction

Kayla E. Scharfstein, Arun Kumar Kuchibhotla

Abstract

Given that machine learning algorithms are increasingly being deployed to aid in high stakes decision-making, uncertainty quantification methods that wrap around these black box models such as conformal prediction have received much attention in recent years. In sequential settings, where data are observed/generated in a streaming fashion, traditional conformal methods do not provide any guarantee without fixing the sample size. More importantly, traditional conformal methods cannot cope with sequentially updated predictions. As such, we develop an extension of the conformal prediction and related probably approximately correct (PAC) prediction frameworks to sequential settings where the number of data points is not fixed in advance. The resulting prediction sets are anytime-valid in that their expected coverage is at the required level at any time chosen by the analyst even if this choice depends on the data. We present theoretical guarantees for our proposed methods and demonstrate their validity and utility on simulated and real datasets.

Author abstract on arXiv

Description

A prediction interval calibrated for a fixed sample size can lose its guarantee if an analyst decides when to stop from the incoming data. These sequential procedures allow predictions to be updated and evaluated at a data-dependent time.

Recent talks

All talks

Mean-Median gaps in real line and Inference in Infinite-dimensions

Statistical Foundations of AI, UC Santa Barbara

Statistical Inference for Linear Programming Problems: Debiasing and Regularity

ASA Statistical Learning and Data Science

Adaptive and Asymptotically Valid Inference for Regularized Problems

Workshop on Selective Inference, Indian School of Business, Hyderabad

Some comments on Inference under Irregularity

Statistics Seminar, George Mason University

Multiply Robust Prediction Sets for Missing Data

Joint Statistical Meetings, Invited Session

Background

I joined Carnegie Mellon in 2020 after completing my PhD in Statistics at the Wharton School under the guidance of Lawrence D. Brown and Andreas Buja. I earned my bachelor’s and master’s degrees in Statistics at the Indian Statistical Institute.

In 2026, I received an Alfred P. Sloan Research Fellowship and the Gottfried E. Noether Early Career Scholar Award.

Collaborators