arXiv subjects

Tzviel Frostig

Publications and source records attributed to Tzviel Frostig.

7 recordsLinked to original sources

arXiv2026

Extending Item Response Theory for Efficient and Meaningful Multilingual Evaluation

Multilingual benchmarks are central to evaluating large language models (LLMs) across languages, but they suffer from three issues: exhaustive evaluation scales linearly with the number of languages, automatic translation introduces errors that are easily missed at scale, and some items conflate general and culture-specific knowledge. We address all three with a unified statistical framework, Multilingual-IRT, which extends Item Response Theory with per-language difficulty deviations, split discriminability separating content from language effects, and per-language ability residuals. Fitting Multilingual-IRT on 25 LLMs across 29 languages of MMLU-Pro-X, we show that its fitted parameters support three practical applications: predicting unobserved (item, LLM, language) instances with 11-16% lower binary cross-entropy than the strongest accuracy-based baseline, surfacing candidate translation errors distributed across all 28 non-English languages, whereas accuracy-based baselines concentrate detections in a few languages, and recovering culture-specific items that accuracy-based baselines miss.

cs.CL↗

arXiv2025

Direction Preferring Confidence Intervals

Confidence intervals (CIs) are instrumental in statistical analysis, providing a range estimate of the parameters. In modern statistics, selective inference is common, where only certain parameters are highlighted. However, this selective approach can bias the inference, leading some to advocate for the use of CIs over p-values. To increase the flexibility of confidence intervals, we introduce direction-preferring CIs, enabling analysts to focus on parameters trending in a particular direction. We present these types of CIs in two settings: First, when there is no selection of parameters; and second, for situations involving parameter selection, where we offer a conditional version of the direction-preferring CIs. Both of these methods build upon the foundations of Modified Pratt CIs, which rely on non-equivariant acceptance regions to achieve longer intervals in exchange for improved sign exclusions. We show that for selected parameters out of m > 1 initial parameters of interest, CIs aimed at controlling the false coverage rate, have higher power to determine the sign compared to conditional CIs. We also show that conditional confidence intervals control the marginal false coverage rate (mFCR) under any dependency.

stat.ME↗

arXiv2024

Robust CATE Estimation Using Novel Ensemble Methods

The estimation of Conditional Average Treatment Effects (CATE) is crucial for understanding the heterogeneity of treatment effects in clinical trials. We evaluate the performance of common methods, including causal forests and various meta-learners, across a diverse set of scenarios, revealing that each of the methods struggles in one or more of the tested scenarios. Given the inherent uncertainty of the data-generating process in real-life scenarios, the robustness of a CATE estimator to various scenarios is critical for its reliability. To address this limitation of existing methods, we propose two new ensemble methods that integrate multiple estimators to enhance prediction stability and performance - Stacked X-Learner which uses the X-Learner with model stacking for estimating the nuisance functions, and Consensus Based Averaging (CBA), which averages only the models with highest internal agreement. We show that these models achieve good performance across a wide range of scenarios varying in complexity, sample size and structure of the underlying-mechanism, including a biologically driven model for PD-L1 inhibition pathway for cancer treatment. Furthermore, we demonstrate improved performance by the Stacked X-Learner also when comparing to other ensemble methods, including R-Stacking, Causal-Stacking and others.

stat.ME↗

arXiv2024

Causal Responder Detection

We introduce the causal responders detection (CARD), a novel method for responder analysis that identifies treated subjects who significantly respond to a treatment. Leveraging recent advances in conformal prediction, CARD employs machine learning techniques to accurately identify responders while controlling the false discovery rate in finite sample sizes. Additionally, we incorporate a propensity score adjustment to mitigate bias arising from non-random treatment allocation, enhancing the robustness of our method in observational settings. Simulation studies demonstrate that CARD effectively detects responders with high power in diverse scenarios.

stat.ME↗

arXiv2022

Inferring on joint associations from marginal associations and a reference sample

We present a method to infer on joint regression coefficients obtained from marginal regressions using a reference panel. This type of scenario is common in genetic fine-mapping, where the estimated marginal associations are reported in genomewide association studies (GWAS), and a reference panel is used for inference on the association in a joint regression model. We show that ignoring the uncertainty due to the use of a reference panel instead of the original design matrix, can lead to a severe inflation of false discoveries and a lack of replicable findings. We derive the asymptotic distribution of the estimated coefficients in the joint regression model, and show how it can be used to produce valid inference. We address two settings: inference within regions that are pre-selected, as well as within regions that are selected based on the same data. By means of real data examples and simulations we demonstrate the usefulness of our suggested methodology.

stat.ME↗

arXiv2022

A statistical framework for efficient out of distribution detection in deep neural networks

Background. Commonly, Deep Neural Networks (DNNs) generalize well on samples drawn from a distribution similar to that of the training set. However, DNNs' predictions are brittle and unreliable when the test samples are drawn from a dissimilar distribution. This is a major concern for deployment in real-world applications, where such behavior may come at a considerable cost, such as industrial production lines, autonomous vehicles, or healthcare applications. Contributions. We frame Out Of Distribution (OOD) detection in DNNs as a statistical hypothesis testing problem. Tests generated within our proposed framework combine evidence from the entire network. Unlike previous OOD detection heuristics, this framework returns a $p$-value for each test sample. It is guaranteed to maintain the Type I Error (T1E - incorrectly predicting OOD for an actual in-distribution sample) for test data. Moreover, this allows to combine several detectors while maintaining the T1E. Building on this framework, we suggest a novel OOD procedure based on low-order statistics. Our method achieves comparable or better results than state-of-the-art methods on well-accepted OOD benchmarks, without retraining the network parameters or assuming prior knowledge on the test distribution -- and at a fraction of the computational cost.

cs.LG↗

arXiv2019

Testing the equality of multivariate means when $p>n$ by combining the Hoteling and Simes tests

We propose a method of testing the shift between mean vectors of two multivariate Gaussian random variables in a high-dimensional setting incorporating the possible dependency and allowing $p > n$. This method is a combination of two well-known tests: the Hotelling test and the Simes test. The tests are integrated by sampling several dimensions at each iteration, testing each using the Hotelling test, and combining their results using the Simes test. We prove that this procedure is valid asymptotically. This procedure can be extended to handle non-equal covariance matrices by plugging in the appropriate extension of the Hotelling test. Using a simulation study, we show that the proposed test is advantageous over state-of-the-art tests in many scenarios and robust to violation of the Gaussian assumption.

stat.ME↗