Julia Nakhleh

jnakhleh@wisc.edu

I'm a PhD candidate at the University of Wisconsin-Madison, advised by Robert D. Nowak. My interests broadly lie in the mathematical theory and foundations of neural networks and of machine learning in general, as well as connections with applied and computational harmonic analysis, nonparametric regression/function estimation, and compressed sensing.

My recent research projects (see Publications) have theoretically characterized the functional properties and sparsity of neural networks that are globally optimal for various weight-regularized training problems. I have also previously done applied work on out-of-distribution (OOD) detection and machine learning for physics applications.

Publications

Julia B. Nakhleh and Robert D. Nowak. "Global Minimizers of $\ell^p$ Regularized Objectives Yield the Sparsest ReLU Networks." NeurIPS 2025 (accepted). [arXiv]

Julia B. Nakhleh, Joseph Shenouda, and Robert D. Nowak. "A New Neural Kernel Regime: the Inductive Bias of Multi-Task Learning." NeurIPS 2024. [proceedings] [arXiv]

Julian J. Katz-Samuels*, Julia B. Nakhleh*, Robert D. Nowak, and Yixuan Li. "Training OOD Detectors in their Natural Habitats." ICML 2022. [proceedings] [arXiv] *equal contribution

M. Giselle Fernández-Godino, Michael J. Grosskopf, Julia B. Nakhleh, Brandon M. Wilson, John L. Kline, and Gowri Srinivasan. "Identifying Entangled Physics Relationships through Sparse Matrix Decomposition to Inform Plasma Fusion Design." IEEE Transactions on Plasma Science 2021. [IEEExplore] [arXiv]

Julia B. Nakhleh, M. Giselle Fernández-Godino, Michael J. Grosskopf, Brandon M. Wilson, John L. Kline, and Gowri Srinivasan. "Exploring Sensitivity of ICF Outputs to Design Parameters in Experiments using Machine Learning." IEEE Transactions on Plasma Science 2021. [IEEExplore] [arXiv]