arXiv subjects

Yuval Benjamini

Publications and source records attributed to Yuval Benjamini.

13 recordsLinked to original sources

arXiv2026

Quantifying Ranking Uncertainty in LLM Benchmarks

Pretrained models are typically ranked on multi-task leaderboards to assess their effectiveness across diverse tasks. Rank confidence intervals were recently introduced as a method to quantify the uncertainty in these rankings by aggregating pairwise hypothesis tests. In this work, we analyze the sources of uncertainty in the knowledge evaluation benchmark MMLU and show how hypothesis tests can be modified to account for their effects. We demonstrate that ranking variability across MMLU subjects is substantial and should be considered when comparing LLMs or identifying the top-performing models.

cs.LG↗

arXiv2026

Rank Intervals for Leaderboards: A Hierarchical Framework for Model Evaluation

Pretrained models are often evaluated on multi-task leaderboards to measure their applicability in diverse contexts. However, current methods for aggregating performance across tasks into leaderboard-level rankings do not address the uncertainty and variability at the task level. While recent works have proposed interval-based model rankings, the principled aggregation of uncertainty from individual tasks to leaderboard-level rankings remains unaddressed, and variation in models' performance across tasks is frequently obscured. In this work, we introduce a hierarchical framework that constructs model rank intervals with statistical guarantees at both levels: task-level rank confidence intervals from pairwise comparisons, and leaderboard-level rank prediction intervals using a conformal approach. This enables reliable quantification of model rank for each observed task and for new potential tasks. Experiments on simulated data and the TabArena and PromptEval (MMLU) benchmarks show that our method yields statistically valid and informative intervals, enabling reliable, uncertainty-aware model ranking on leaderboards.

stat.ML↗

arXiv2025

CardiCat: a Variational Autoencoder for High-Cardinality Tabular Data

High-cardinality categorical features are a common characteristic of mixed-type tabular datasets. Existing generative model architectures struggle to learn the complexities of such data at scale, primarily due to the difficulty of parameterizing the categorical features. In this paper, we present a general variational autoencoder model, CardiCat, that can accurately fit imbalanced high-cardinality and heterogeneous tabular data. Our method substitutes one-hot encoding with regularized dual encoder-decoder embedding layers, which are jointly learned. This approach enables us to use embeddings that depend also on the other covariates, leading to a compact and homogenized parameterization of categorical features. Our model employs a considerably smaller trainable parameter space than competing methods, enabling learning at a large scale. CardiCat generates high-quality synthetic data that better represent high-cardinality and imbalanced features compared to competing VAE models for multiple real and simulated datasets.

cs.LG↗

arXiv2023

Class Distribution Shifts in Zero-Shot Learning: Learning Robust Representations

Zero-shot learning methods typically assume that the new, unseen classes encountered during deployment come from the same distribution as the the classes in the training set. However, real-world scenarios often involve class distribution shifts (e.g., in age or gender for person identification), posing challenges for zero-shot classifiers that rely on learned representations from training classes. In this work, we propose and analyze a model that assumes that the attribute responsible for the shift is unknown in advance. We show that in this setting, standard training may lead to non-robust representations. To mitigate this, we develop an algorithm for learning robust representations in which (a) synthetic data environments are constructed via hierarchical sampling, and (b) environment balancing penalization, inspired by out-of-distribution problems, is applied. We show that our algorithm improves generalization to diverse class distributions in both simulations and experiments on real-world datasets.

cs.LG↗

arXiv2023

Confident Feature Ranking

Machine learning models are widely applied in various fields. Stakeholders often use post-hoc feature importance methods to better understand the input features' contribution to the models' predictions. The interpretation of the importance values provided by these methods is frequently based on the relative order of the features (their ranking) rather than the importance values themselves. Since the order may be unstable, we present a framework for quantifying the uncertainty in global importance values. We propose a novel method for the post-hoc interpretation of feature importance values that is based on the framework and pairwise comparisons of the feature importance values. This method produces simultaneous confidence intervals for the features' ranks, which include the ``true'' (infinite sample) ranks with high probability, and enables the selection of the set of the top-k important features.

stat.ML↗

arXiv2023

Logistic Regression Equivalence: A Framework for Comparing Logistic Regression Models Across Populations

In this paper we discuss how to evaluate the differences between fitted logistic regression models across sub-populations. Our motivating example is in studying computerized diagnosis for learning disabilities, where sub-populations based on gender may or may not require separate models. In this context, significance tests for hypotheses of no difference between populations may provide perverse incentives, as larger variances and smaller samples increase the probability of not-rejecting the null. We argue that equivalence testing for a prespecified tolerance level on population differences incentivizes accuracy in the inference. We develop a cascading set of equivalence tests, in which each test addresses a different aspect of the model: the way the phenomenon is coded in the regression coefficients, the individual predictions in the per example log odds ratio and the overall accuracy in the mean square prediction error. For each equivalence test, we propose a strategy for setting the equivalence thresholds. The large-sample approximations are validated using simulations. For diagnosis data, we show examples for equivalent and non-equivalent models.

stat.ME↗

arXiv2021

Detecting Differences Between Correlation-Matrix Populations due to Single-variable Perturbations, with Application to Resting State fMRI

Correlation matrices are widely used to analyze the interdependence of variables in various real-world scenarios. Often, a perturbation in a few variables leads to mild differences in many correlation coefficients associated with these variables. We propose an efficient low-dimensional model that characterizes these differences as a product of single-variable effects. We develop methods for point estimation, confidence intervals, and hypothesis testing for this model. Importantly, our methods can account for both the variability in individual correlation matrices and for within-group variability. In simulations, our model shows increased power compared to competing approaches. We use the model to analyze resting-state functional MRI correlation matrices in patients with transient global amnesia and healthy controls. Our model detects significant decreases in synchronization for the patient population in several brain regions, which could not have been detected using previous methods without prior knowledge. Our methods are available in the open-source package \emph{github.com/itamarfaran/corrpops}.

stat.AP↗

arXiv2020

Predicting Classification Accuracy When Adding New Unobserved Classes

Multiclass classifiers are often designed and evaluated only on a sample from the classes on which they will eventually be applied. Hence, their final accuracy remains unknown. In this work we study how a classifier's performance over the initial class sample can be used to extrapolate its expected accuracy on a larger, unobserved set of classes. For this, we define a measure of separation between correct and incorrect classes that is independent of the number of classes: the "reversed ROC" (rROC), which is obtained by replacing the roles of classes and data-points in the common ROC. We show that the classification accuracy is a function of the rROC in multiclass classifiers, for which the learned representation of data from the initial class sample remains unchanged when new classes are added. Using these results we formulate a robust neural-network-based algorithm, "CleaneX", which learns to estimate the accuracy of such classifiers on arbitrarily large sets of classes. Unlike previous methods, our method uses both the observed accuracies of the classifier and densities of classification scores, and therefore achieves remarkably better predictions than current state-of-the-art methods on both simulations and real datasets of object detection, face recognition, and brain decoding.

cs.LG↗

arXiv2017

Extrapolating Expected Accuracies for Large Multi-Class Problems

The difficulty of multi-class classification generally increases with the number of classes. Using data from a subset of the classes, can we predict how well a classifier will scale with an increased number of classes? Under the assumptions that the classes are sampled identically and independently from a population, and that the classifier is based on independently learned scoring functions, we show that the expected accuracy when the classifier is trained on k classes is the (k-1)st moment of a certain distribution that can be estimated from data. We present an unbiased estimation method based on the theory, and demonstrate its application on a facial recognition example.

stat.ML↗

arXiv2017

Better-Than-Chance Classification for Signal Detection

The estimated accuracy of a classifier is a random quantity with variability. A common practice in supervised machine learning, is thus to test if the estimated accuracy is significantly better than chance level. This method of signal detection is particularly popular in neuroimaging and genetics. We provide evidence that using a classifier's accuracy as a test statistic can be an underpowered strategy for finding differences between populations, compared to a bona-fide statistical test. It is also computationally more demanding than a statistical test. Via simulation, we compare test statistics that are based on classification accuracy, to others based on multivariate test statistics. We find that probability of detecting differences between two distributions is lower for accuracy based statistics. We examine several candidate causes for the low power of accuracy tests. These causes include: the discrete nature of the accuracy test statistic, the type of signal accuracy tests are designed to detect, their inefficient use of the data, and their regularization. When the purposes of the analysis is not signal detection, but rather, the evaluation of a particular classifier, we suggest several improvements to increase power. In particular, to replace V-fold cross validation with the Leave-One-Out Bootstrap.

stat.ME↗

arXiv2016

Estimating mutual information in high dimensions via classification error

Multivariate pattern analyses approaches in neuroimaging are fundamentally concerned with investigating the quantity and type of information processed by various regions of the human brain; typically, estimates of classification accuracy are used to quantify information. While a extensive and powerful library of methods can be applied to train and assess classifiers, it is not always clear how to use the resulting measures of classification performance to draw scientific conclusions: e.g. for the purpose of evaluating redundancy between brain regions. An additional confound for interpreting classification performance is the dependence of the error rate on the number and choice of distinct classes obtained for the classification task. In contrast, mutual information is a quantity defined independently of the experimental design, and has ideal properties for comparative analyses. Unfortunately, estimating the mutual information based on observations becomes statistically infeasible in high dimensions without some kind of assumption or prior. In this paper, we construct a novel classification-based estimator of mutual information based on high-dimensional asymptotics. We show that in a particular limiting regime, the mutual information is an invertible function of the expected $k$-class Bayes error. While the theory is based on a large-sample, high-dimensional limit, we demonstrate through simulations that our proposed estimator has superior performance to the alternatives in problems of moderate dimensionality.

stat.ML↗

arXiv2016

How many faces can be recognized? Performance extrapolation for multi-class classification

The difficulty of multi-class classification generally increases with the number of classes. Using data from a subset of the classes, can we predict how well a classifier will scale with an increased number of classes? Under the assumption that the classes are sampled exchangeably, and under the assumption that the classifier is generative (e.g. QDA or Naive Bayes), we show that the expected accuracy when the classifier is trained on $k$ classes is the $k-1$st moment of a \emph{conditional accuracy distribution}, which can be estimated from data. This provides the theoretical foundation for performance extrapolation based on pseudolikelihood, unbiased estimation, and high-dimensional asymptotics. We investigate the robustness of our methods to non-generative classifiers in simulations and one optical character recognition example.

stat.ML↗

arXiv2014

The shuffle estimator for explainable variance in fMRI experiments

In computational neuroscience, it is important to estimate well the proportion of signal variance in the total variance of neural activity measurements. This explainable variance measure helps neuroscientists assess the adequacy of predictive models that describe how images are encoded in the brain. Complicating the estimation problem are strong noise correlations, which may confound the neural responses corresponding to the stimuli. If not properly taken into account, the correlations could inflate the explainable variance estimates and suggest false possible prediction accuracies. We propose a novel method to estimate the explainable variance in functional MRI (fMRI) brain activity measurements when there are strong correlations in the noise. Our shuffle estimator is nonparametric, unbiased, and built upon the random effect model reflecting the randomization in the fMRI data collection process. Leveraging symmetries in the measurements, our estimator is obtained by appropriately permuting the measurement vector in such a way that the noise covariance structure is intact but the explainable variance is changed after the permutation. This difference is then used to estimate the explainable variance. We validate the properties of the proposed method in simulation experiments. For the image-fMRI data, we show that the shuffle estimates can explain the variation in prediction accuracy for voxels within the primary visual cortex (V1) better than alternative parametric methods.

stat.AP↗