arXiv subjects

Mohamed Salem

Publications and source records attributed to Mohamed Salem.

3 recordsLinked to original sources

arXiv2026

Valid Feature-Level Inference for Tabular Foundation Models via the Conditional Randomization Test

Modern machine learning models are highly expressive but notoriously difficult to analyze statistically. In particular, while black-box predictors can achieve strong empirical performance, they rarely provide valid hypothesis tests or p-values for assessing whether individual features contain information about a target variable. This article presents a practical approach to feature-level hypothesis testing that combines the Conditional Randomization Test (CRT) with TabPFN, a probabilistic foundation model for tabular data. The resulting procedure yields finite-sample valid p-values for conditional feature relevance, even in nonlinear and correlated settings, without requiring model retraining or parametric assumptions.

cs.LG↗

arXiv2025

Likelihood-guided Regularization in Attention Based Models

The transformer architecture has demonstrated strong performance in classification tasks involving structured and high-dimensional data. However, its success often hinges on large- scale training data and careful regularization to prevent overfitting. In this paper, we intro- duce a novel likelihood-guided variational Ising-based regularization framework for Vision Transformers (ViTs), which simultaneously enhances model generalization and dynamically prunes redundant parameters. The proposed variational Ising-based regularization approach leverages Bayesian sparsification techniques to impose structured sparsity on model weights, allowing for adaptive architecture search during training. Unlike traditional dropout-based methods, which enforce fixed sparsity patterns, the variational Ising-based regularization method learns task-adaptive regularization, improving both efficiency and interpretability. We evaluate our approach on benchmark vision datasets, including MNIST, Fashion-MNIST, CIFAR-10, and CIFAR-100, demonstrating improved generalization under sparse, complex data and allowing for principled uncertainty quantification on both weights and selection parameters. Additionally, we show that the Ising regularizer leads to better-calibrated probability estimates and structured feature selection through uncertainty-aware attention mechanisms. Our results highlight the effectiveness of structured Bayesian sparsification in enhancing transformer-based architectures, offering a principled alternative to standard regularization techniques.

stat.ML↗

arXiv2016

Intra-wire coupling in segmented Ni/Cu nanowires deposited by electrodeposition

Segmented magnetic nanowires are a promising route for the development of three dimensional data storage techniques. Such devices require a control of the coercive field and the coupling mechanisms between individual magnetic elements. In our study, we investigate electrodeposited nanomagnets within host templates using vibrating sample magnetometry and observe a strong dependence between nanowire length and coercive field (25 nm to 5 $μ$m) and diameter (25 nm to 45 nm). A transition from a magnetization reversal through coherent rotation to domain wall propagation is observed at an aspect ratio of approximately 2. Our results are further reinforced via micromagnetic simulations and angle dependent hysteresis loops. The found behavior is exploited to create nanowires consisting of a fixed and a free segment in a spin-valve like structure. The wires are released from the membrane and electrically contacted, displaying a giant magnetoresistance effect that is attributed to individual switching of the coupled nanomagnets. We develop a simple analytical model to describe the observed switching phenomena and to predict stable and unstable regimes in coupled nanomagnets of certain geometries.

cond-mat.mtrl-sci↗