Searcharxiv⌕ Search

arXiv subjects

Etienne Roquain

Publications and source records attributed to Etienne Roquain.

At least 19 recordsLinked to original sources
arXiv2026

Online selective conformal inference: adaptive scores, convergence rates and optimality

In a supervised online setting, quantifying uncertainty has been proposed in the seminal work of Gibbs and Candès (2021). For any given point-prediction algorithm, their method (ACI) produces a conformal prediction set with an average miscoverage getting close to a prespecified level $α$ for a long time horizon. We introduce an extended version of this algorithm, called OnlineSCI, allowing the user to additionally select times where such an inference should be made. OnlineSCI encompasses several prominent online selective tasks, such as building prediction intervals for extreme outcomes, classification with abstention, and online testing. OnlineSCI controls the false coverage proportion among selected times via a pathwise bound for arbitrary sequences, as well as the instantaneous error rate conditional on selection, up to a non-asymptotic remainder term, under stochastic assumptions. Importantly, our theory covers the case where OnlineSCI updates the point-prediction algorithm at each time step, a property which we refer to as {\it adaptive} capability. We show that the adaptive versions of OnlineSCI can converge to an optimal solution and provide an explicit convergence rate in several application settings, under model- and estimator-specific regularity conditions. The favorable behavior of OnlineSCI in practice is illustrated by numerical experiments.

math.ST↗
arXiv2026

Reliable conformal novelty detection at the decision boundary

Novelty detection via conformal $p$-values and BH procedure provides distribution-free global false discovery rate (FDR) control. We present here fundamental limits of this approach by showing that it does not produce reliable detection at the decision boundary. We study boundary false discovery rate (bFDR), the probability that the least extreme reported novelty is in fact a null observation. We first show that the support line (SL) procedure, controlling the bFDR in the continuous independent framework, fails to control the bFDR in the conformal case. We then present several modifications of the SL procedure that restore reliability at the decision boundary, by controlling the bFDR, with specific improvements in situations where many novelties are expected (adaptive procedures) and when the calibration sample is too small with respect to the test sample (subsampled procedures). Numerical experiments with both synthetic and real data support our findings and show the relevance of the new proposed approach.

stat.ME↗
arXiv2026

Selecting Informative Conformal Prediction Sets with an Optimized FCR-Controlled Approach

Conformal methods provide prediction sets for outcomes with confidence guarantees. We study their use in a selective inference setting, where inference is performed only when the prediction set is informative. The analyst may consider as informative, for example, cases with prediction sets that are sufficiently small, exclude null values, or satisfy other appropriate monotone constraints. Because inference is typically restricted to informative cases in practical applications, accounting for the resulting selection bias is crucial to maintaining false coverage rate (FCR) control. A general framework for constructing such informative conformal prediction sets while controlling the FCR on the selected sample was suggested in Gazin et al. (2025). In this work we focus on oracle-guided procedures. We derive the optimal decision policy under a suitable power objective in the oracle setting where the probability of belonging to each prediction set can be computed. In practice, of course, only estimated probabilities are available. We therefore introduce a calibration procedure that adjusts the oracle policy to maintain finite sample FCR control. We show that this approach can achieve substantially higher power than available alternatives. We demonstrate the effectiveness of our new methods for classification outcomes on both real and simulated data.

stat.ME↗
arXiv2026

Confidence envelopes for the false discoveries with heterogeneous data

In the context of selective inference, confidence envelopes for the false discoveries allow the user to select any subset of null hypotheses while having a statistical guarantee on the number of false discoveries in the selected set. Many constructions of such envelopes have been proposed recently, using local test families (Genovese and Wasserman, 2006; Goeman and Solari, 2011), paths (Katsevich and Ramdas, 2020) or interpolation (Blanchard et al., 2020a). All those methods have in common that they have been well-studied for the homogeneous case where all p-values under the null have a uniform distribution over [0, 1]. However, in many applications the data are heterogeneous and discrete, hence the p-values have heterogeneous, discrete distributions, and the previous constructions may incur a loss of power, in the sense that they over-estimate the number of false discoveries. In this paper, we bridge the previous constructions under the homogeneous case with new tools. We also apply these tools to propose several confidence envelopes based on tools tailored for heterogeneous data, like the Bretagnolle inequality, or a new variant of the Simes inequality. We compare these new envelopes to their homogeneous counterparts on simulated data.

math.ST↗
arXiv2025

Powerful batch conformal prediction for classification

In a split conformal framework with $K$ classes, a calibration sample of $n$ labeled examples is observed for inference on the label of a new unlabeled example. We explore the setting where a `batch' of $m$ independent such unlabeled examples is given, and the goal is to construct a batch prediction set with 1-$α$ coverage. Unlike individual prediction sets, the batch prediction set is a collection of label vectors of size $m$, while the calibration sample consists of univariate labels. A natural approach is to apply the Bonferroni correction, which concatenates individual prediction sets at level $1-α/m$. We propose a uniformly more powerful solution, based on specific combinations of conformal $p$-values that exploit the Simes inequality. We provide a general recipe for valid inference with any combinations of conformal $p$-values, and compare the performance of several useful choices. Intuitively, the pooled evidence of relatively `easy' examples within the batch can help provide narrower batch prediction sets. Additionally, we introduce a more computationally intensive method that aggregates batch scores and can be even more powerful. The theoretical guarantees are established when all examples are independent and identically distributed (iid), as well as more generally when iid is assumed only conditionally within each class. Notably, our results remain valid under label distribution shift, since the distribution of the labels need not be the same in the calibration sample and in the new batch. The effectiveness of the methods is highlighted through illustrative synthetic and real data examples.

stat.ME↗
arXiv2024

Selecting informative conformal prediction sets with false coverage rate control

In supervised learning, including regression and classification, conformal methods provide prediction sets for the outcome/label with finite sample coverage for any machine learning predictor. We consider here the case where such prediction sets come after a selection process. The selection process requires that the selected prediction sets be `informative' in a well defined sense. We consider both the classification and regression settings where the analyst may consider as informative only the sample with prediction sets small enough, excluding null values, or obeying other appropriate `monotone' constraints. We develop a unified framework for building such informative conformal prediction sets while controlling the false coverage rate (FCR) on the selected sample. While conformal prediction sets after selection have been the focus of much recent literature in the field, the new introduced procedures, called InfoSP and InfoSCOP, are to our knowledge the first ones providing FCR control for informative prediction sets. We show the usefulness of our resulting procedures on real and simulated data.

math.ST↗
arXiv2024

False discovery proportion envelopes with m-consistency

We provide new non-asymptotic false discovery proportion (FDP) confidence envelopes in several multiple testing settings relevant for modern high dimensional-data methods. We revisit the multiple testing scenarios considered in the recent work of Katsevich and Ramdas (2020): top-$k$, preordered (including knockoffs), online. Our emphasis is on obtaining FDP confidence bounds that both have non-asymptotic coverage and are asymptotically accurate in a specific sense, as the number $m$ of tested hypotheses grows. Namely, we introduce and study the property (which we call $m$-consistency) that the confidence bound converges to or below the desired level $α$ when applied to a specific reference $α$-level false discovery rate (FDR) controlling procedure. In this perspective, we derive new bounds that provide improvements over existing ones, both theoretically and practically, and are suitable for situations where at least a moderate number of rejections is expected. These improvements are illustrated with numerical experiments and real data examples. In particular, the improvement is significant in the knockoffs setting, which shows the impact of the method for a practical use. As side results, we introduce a new confidence envelope for the empirical cumulative distribution function of i.i.d. uniform variables, and we provide new power results in sparse cases, both being of independent interest.

math.ST↗
arXiv2024

Transductive conformal inference with adaptive scores

Conformal inference is a fundamental and versatile tool that provides distribution-free guarantees for many machine learning tasks. We consider the transductive setting, where decisions are made on a test sample of $m$ new points, giving rise to $m$ conformal $p$-values. While classical results only concern their marginal distribution, we show that their joint distribution follows a Pólya urn model, and establish a concentration inequality for their empirical distribution function. The results hold for arbitrary exchangeable scores, including adaptive ones that can use the covariates of the test+calibration samples at training stage for increased accuracy. We demonstrate the usefulness of these theoretical results through uniform, in-probability guarantees for two machine learning tasks of current interest: interval prediction for transductive transfer learning and novelty detection based on two-class classification.

stat.ME↗
arXiv2023

False membership rate control in mixture models

The clustering task consists in partitioning elements of a sample into homogeneous groups. Most datasets contain individuals that are ambiguous and intrinsically difficult to attribute to one or another cluster. However, in practical applications, misclassifying individuals is potentially disastrous and should be avoided. To keep the misclassification rate small, one can decide to classify only a part of the sample. In the supervised setting, this approach is well known and referred to as classification with an abstention option. In this paper the approach is revisited in an unsupervised mixture model framework and the purpose is to develop a method that comes with the guarantee that the false membership rate (FMR) does not exceed a pre-defined nominal level $α$. A plug-in procedure is proposed, for which a theoretical analysis is provided, by quantifying the FMR deviation with respect to the target level $α$ with explicit remainder terms. Bootstrap versions of the procedure are shown to improve the performance in numerical experiments.

math.ST↗
arXiv2023

Adaptive novelty detection with false discovery rate guarantee

This paper studies the semi-supervised novelty detection problem where a set of "typical" measurements is available to the researcher. Motivated by recent advances in multiple testing and conformal inference, we propose AdaDetect, a flexible method that is able to wrap around any probabilistic classification algorithm and control the false discovery rate (FDR) on detected novelties in finite samples without any distributional assumption other than exchangeability. In contrast to classical FDR-controlling procedures that are often committed to a pre-specified p-value function, AdaDetect learns the transformation in a data-adaptive manner to focus the power on the directions that distinguish between inliers and outliers. Inspired by the multiple testing literature, we further propose variants of AdaDetect that are adaptive to the proportion of nulls while maintaining the finite-sample FDR control. The methods are illustrated on synthetic datasets and real-world datasets, including an application in astrophysics.

stat.ME↗
arXiv2023

Sharp multiple testing boundary for sparse sequences

This work investigates multiple testing by considering minimax separation rates in the sparse sequence model, when the testing risk is measured as the sum FDR+FNR (False Discovery Rate plus False Negative Rate). First using the popular beta-min separation condition, with all nonzero signals separated from $0$ by at least some amount, we determine the sharp minimax testing risk asymptotically and thereby explicitly describe the transition from "achievable multiple testing with vanishing risk" to "impossible multiple testing". Adaptive multiple testing procedures achieving the corresponding optimal boundary are provided: the Benjamini--Hochberg procedure with a properly tuned level, and an empirical Bayes $\ell$-value (`local FDR') procedure. We prove that the FDR and FNR make non-symmetric contributions to the testing risk for most optimal procedures, the FNR part being dominant at the boundary. The multiple testing hardness is then investigated for classes of arbitrary sparse signals. A number of extensions, including results for classification losses and convergence rates in the case of large signals, are also investigated.

math.ST↗
arXiv2023

Online multiple testing with super-uniformity reward

Valid online inference is an important problem in contemporary multiple testing research,to which various solutions have been proposed recently. It is well-known that these existing methods can suffer from a significant loss of power if the null $p$-values are conservative. In this work, we extend the previously introduced methodology to obtain more powerful procedures for the case of super-uniformly distributed $p$-values. These types of $p$-values arise in important settings, e.g. when discrete hypothesis tests are performed or when the $p$-values are weighted. To this end, we introduce the method of super-uniformity reward (SUR) that incorporates information about the individual null cumulative distribution functions. Our approach yields several new 'rewarded' procedures that offer uniform power improvements over known procedures and come with mathematical guarantees for controlling online error criteria based either on the family-wise error rate (FWER) or the marginal false discovery rate (mFDR). We illustrate the benefit of super-uniform rewarding in real-data analyses and simulation studies. While discrete tests serve as our leading example, we also show how our method can be applied to weighted $p$-values.

stat.ME↗
arXiv2022

Empirical Bayes cumulative $\ell$-value multiple testing procedure for sparse sequences

In the sparse sequence model, we consider a popular Bayesian multiple testing procedure and investigate for the first time its behaviour from the frequentist point of view. Given a spike-and-slab prior on the high-dimensional sparse unknown parameter, one can easily compute posterior probabilities of coming from the spike, which correspond to the well known local-fdr values, also called $\ell$-values. The spike-and-slab weight parameter is calibrated in an empirical Bayes fashion, using marginal maximum likelihood. The multiple testing procedure under study, called here the cumulative $\ell$-value procedure, ranks coordinates according to their empirical $\ell$-values and thresholds so that the cumulative ranked sum does not exceed a user-specified level $t$. We validate the use of this method from the multiple testing perspective: for alternatives of appropriately large signal strength, the false discovery rate (FDR) of the procedure is shown to converge to the target level $t$, while its false negative rate (FNR) goes to $0$. We complement this study by providing convergence rates for the method. Additionally, we prove that the $q$-value multiple testing procedure shares similar convergence rates in this model.

math.ST↗
arXiv2021

Semi-supervised multiple testing

An important limitation of standard multiple testing procedures is that the null distribution should be known. Here, we consider a null distribution-free approach for multiple testing in the following semi-supervised setting: the user does not know the null distribution, but has at hand a sample drawn from this null distribution. In practical situations, this null training sample (NTS) can come from previous experiments, from a part of the data under test, from specific simulations, or from a sampling process. In this work, we present theoretical results that handle such a framework, with a focus on the false discovery rate (FDR) control and the Benjamini-Hochberg (BH) procedure. First, we provide upper and lower bounds for the FDR of the BH procedure based on empirical $p$-values. These bounds match when $α(n+1)/m$ is an integer, where $n$ is the NTS sample size and $m$ is the number of tests. Second, we give a power analysis for that procedure suggesting that the price to pay for ignoring the null distribution is low when $n$ is sufficiently large in front of $m$; namely $n\gtrsim m/(\max(1,k))$, where $k$ denotes the number of ``detectable'' alternatives. Third, to complete the picture, we also present a negative result that evidences an intrinsic transition phase to the general semi-supervised multiple testing problem {and shows that the empirical BH method is optimal in the sense that its performance boundary follows this transition phase}. Our theoretical properties are supported by numerical experiments, which also show that the delineated boundary is of correct order without further tuning any constant. Finally, we demonstrate that our work provides a theoretical ground for standard practice in astronomical data analysis, and in particular for the procedure proposed in \cite{Origin2020} for galaxy detection.

math.ST↗
arXiv2021

Post hoc false discovery proportion inference under a Hidden Markov Model

We address the multiple testing problem under the assumption that the true/false hypotheses are driven by a Hidden Markov Model (HMM), which is recognized as a fundamental setting to model multiple testing under dependence since the seminal work of \citet{sun2009large}. While previous work has concentrated on deriving specific procedures with a controlled False Discovery Rate (FDR) under this model, following a recent trend in selective inference, we consider the problem of establishing confidence bounds on the false discovery proportion (FDP), for a user-selected set of hypotheses that can depend on the observed data in an arbitrary way. We develop a methodology to construct such confidence bounds first when the HMM model is known, then when its parameters are unknown and estimated, including the data distribution under the null and the alternative, using a nonparametric approach. In the latter case, we propose a bootstrap-based methodology to take into account the effect of parameter estimation error. We show that taking advantage of the assumed HMM structure allows for a substantial improvement of confidence bound sharpness over existing agnostic (structure-free) methods, as witnessed both via numerical experiments and real data examples.

stat.ME↗
arXiv2020

False discovery rate control with unknown null distribution: is it possible to mimic the oracle?

Classical multiple testing theory prescribes the null distribution, which is often a too stringent assumption for nowadays large scale experiments. This paper presents theoretical foundations to understand the limitations caused by ignoring the null distribution, and how it can be properly learned from the (same) data-set, when possible. We explore this issue in the case where the null distributions are Gaussian with an unknown rescaling parameters (mean and variance) and the alternative distribution is let arbitrary. While an oracle procedure in that case is the Benjamini Hochberg procedure applied with the true (unknown) null distribution, we pursue the aim of building a procedure that asymptotically mimics the performance of the oracle (AMO in short). Our main result states that an AMO procedure exists if and only if the sparsity parameter $k$ (number of false nulls) is of order less than $n/\log(n)$, where $n$ is the total number of tests. Further sparsity boundaries are derived for general location models where the shape of the null distribution is not necessarily Gaussian. Given our impossibility results, we also pursue a weaker objective, which is to find a confidence region for the oracle. To this end, we develop a distribution-dependent confidence region for the null distribution. As practical by-products, this provides a goodness of fit test for the null distribution, as well as a visual method assessing the reliability of empirical null multiple testing methods. Our results are illustrated with numerical experiments and a companion vignette \cite{RVvignette2020}.

math.ST↗
arXiv2019

Controlling false discovery exceedance for heterogeneous tests

Several classical methods exist for controlling the false discovery exceedance (FDX) for large scale multiple testing problems, among them the Lehmann-Romano procedure ([LR] below) and the Guo-Romano procedure ([GR] below). While these two procedures are the most prominent, they were originally designed for homogeneous test statistics, that is, when the null distribution functions of the $p$-values $F_i$, $1\leq i\leq m$, are all equal. In many applications, however, the data are heterogeneous which leads to heterogeneous null distribution functions. Ignoring this heterogeneity usually induces a conservativeness for the aforementioned procedures. In this paper, we develop three new procedures that incorporate the $F_i$'s, while ensuring the FDX control. The heterogeneous version of [LR], denoted [HLR], is based on the arithmetic average of the $F_i$'s, while the heterogeneous version of [GR], denoted [HGR], is based on the geometric average of the $F_i$'s. We also introduce a procedure [PB], that is based on the Poisson-binomial distribution and that uniformly improves [HLR] and [HGR], at the price of a higher computational complexity. Perhaps surprisingly, this shows that, contrary to the known theory of false discovery rate (FDR) control under heterogeneity, the way to incorporate the $F_i$'s can be particularly simple in the case of FDX control, and does not require any further correction term. The performances of the new proposed procedures are illustrated by real and simulated data in two important heterogeneous settings: first, when the test statistics are continuous but the $p$-values are weighted by some known independent weight vector, e.g., coming from co-data sets; second, when the test statistics are discretely distributed, as is the case for data representing frequencies or counts.

stat.ME↗