About Me

Hello!

I am a PhD Student within the MIND team (former Parietal) at INRIA Saclay, working under the supervision of Alexandre Gramfort and Pedro L. C. Rodrigues on simulation-based inference (SBI) and generative modelling.

The primary focus of my research has been on the use of deep generative models (in particular normalizing flows and diffusion models) to address likelihood-free Bayesian inference (a.k.a. LFI or SBI) tasks and their validation in terms of accuracy, efficiency and reliability. This work is motivated by the goal of efficiently inverting complex simulators from computational neuroscience and more generally to lift the remaining lack of trust for deep generative models to address important questions in experimental sciences. My main projects and collaborations are detailed below.

Before my PhD, I have also had the chance to work more closely on applications in Machine Learning for healthcare at two different start-ups, Covera Health in New York (uncertainty quantification of Deep Learning models on MRIs) and Owkin in Paris (development of Deep Learning models for breast cancer survival analysis on histology images and their calibration).

I am particularly interested in expanding my expertise on generative models and gaining a more profound understanding of their workings, my motivation being a mix of pure curiosity and the goal of applying them to aid experimental sciences accross various biophysical fields.

Here is my CV. For more details, feel free to contact me at julia.linhart@inria.fr .

Main PhD projects and collaborations

• Development of new validation diagnostics for conditional deep generative models [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 Sylvain Le Corff (LPSM - Sorbonne Université) and Gabriel V. Cardoso (CMAP - École Polytechnique).

Funding

My PhD project is part of the ED STIC doctoral school at Université Paris-Saclay. As the first-placed candidate of my year, I am recipient of the Pierre-Aguilar Scholarship of the Capital Fund Management (CFM).

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, Preprint, 2024.