About Me

Hello!

I am a Postdoctoral Researcher at NYU Center for Data Science, working with Julia Kempe (NYU), Shirley Ho (NYU - Flatiron Institute), and Uroš Seljak (UC Berkeley). I completed my PhD at Inria (MIND Team), Université Paris-Saclay under the supervision of Alexandre Gramfort (Meta) and Pedro L. C. Rodrigues (Inria, Grenoble).

My research focuses on probabilistic machine learning, generative modeling and Bayesian inference, with applications across neuroscience, cosmology, and beyond.

I am most excited about how AI can transform the way we do science!

News

Feb 2026: Paper Diffusion Posterior Sampling for SBI in tall data settings accepted at TMLR and awarded with the Journal-to-Conference (J2C) Certification. Will be presented at ICML 2026!
Sep 2025: Started postdoc at NYU Center for Data Science (+ Guest Researcher at the Flatiron Institute).
Aug 2025: New tutorial paper on Simulation-Based Inference is out: link to paper.
~ In collaboration with the Max Planck Institute / University of Tübingen and other main contributors to the sbi Python package.
Dec 2024: Successfully defended my PhD at Inria Saclay: link to thesis, link to video.

Research

A major focus of my research is simulation-based inference (SBI): developing accurate, efficient, and reliable methods for Bayesian inference when the likelihood is intractable but simulations are available. My work combines deep generative models, including normalizing flows and diffusion models, with new sampling strategies and statistical methods for validation and model checking, with a particular interest in robustness to model misspecification.

More broadly, I am interested in understanding how probabilistic ML methods can be reliably used for scientific inference and discovery, particularly when simulations are expensive, models are misspecified, or observations are noisy or corrupted.

Current Directions

My current work focuses on three main directions:

SBI under model misspecification, including Bayesian model checking, posterior predictive checks, and OOD detection;
active learning strategies for simulation-efficient inference;
neural summary statistics for low-budget SBI on weak-lensing maps in cosmology.

Stay tuned…

Main PhD projects

Development of new validation diagnostics for approximate posteriors in neural SBI [1, 2], with an integration to the official sbi python package from the MACKELAB.
Exploration of novel posterior sampling algorithms using deep generative models, for example based on diffusion models when one wishes to condition on multiple observations to get more precise parameter estimations [3]. This is collaborative work with Gabriel V. Cardoso (CMAP - École Polytechnique) and Sylvain Le Corff (LPSM - Sorbonne Université).
Application of SBI to neuroscience time series (EEG) data [4].

Other

Before my PhD, I had the chance to work more closely on ML applications in Medical Imaging at two different start-ups:

Owkin (2021, Paris) - development and calibration of deep survival models for breast cancer prognosis on histology images [5].
Covera Health (2020, New York) - uncertainty quantification of deep classification models on MRIs

Open Source

I am an active contributor to the sbi Python toolbox [6].

CV

Here is my CV. For more details, feel free to contact me at julia.linhart@nyu.edu :)

References

[1] Julia Linhart, Alexandre Gramfort and Pedro L. C. Rodrigues, Validation Diagnostics for SBI algorithms based on Normalizing Flows, ML4PS Workshop, NeurIPS 2022.
[2] Julia Linhart, Alexandre Gramfort and Pedro L. C. Rodrigues, L-C2ST: Local Diagnostics for Posterior Approximations in Simulation-Based Inference, NeurIPS 2023.
[3] Julia Linhart, Gabriel V. Cardoso, Alexandre Gramfort, Sylvain Le Corff and Pedro L. C. Rodrigues, Diffusion posterior sampling for simulation-based inference in tall data settings, TMLR - J2C - ICML 2026.
[4] Julia Linhart, Pedro L. C. Rodrigues, Thomas Moreau, Gilles Louppe, Alexandre Gramfort, Neural Posterior Estimation of hierarchical models in neuroscience, Colloque GRETSI 2022.
[5] I. Garberis, V. Gaury, C. Saillard, et al., Deep learning assessment of metastatic relapse risk from digitized breast cancer histological slides, Nature Communications, 2025.
[6] J. Boelts, M. Deistler, M Gloeckler, et al., sbi reloaded: a toolkit for simulation-based inference workflows, 2024.\