Eduardo Fernandes Montesuma

PhD in mathematical computer science (Université Paris-Saclay & CEA), AI Researcher @ SigmaNova

Optimal Transport | Machine Learning

[Gabriel Peyré @gabrielpeyre]

The alpha version of my new book "Optimal Transport
for Machine Learners" is out, with in particular an online version with interactive figures
https://t.co/xEdZpMXgjx

[Eduardo Fernandes Montesuma @efernand9]

This week, me, @YBendou and @mikegartrell had the chance to present our latest paper, "ReBaPL: Repulsive Bayesian Prompt Learning", at @CVPR

This method is a Bayesian "plug-and-play" on top of existing Prompt Learning methods.

[Eduardo Fernandes Montesuma @efernand9]

Check out our

paper: https://t.co/a5crxBzGPq
code: https://t.co/EyiAg28AGi

[Eduardo Fernandes Montesuma @efernand9]

@YBendou @mikegartrell Check out our paper: https://t.co/yjKzd0tSfa

Code will be released soon!

[Eduardo Fernandes Montesuma @efernand9]

Continuing on the good news, our paper "Wasserstein Gradient Flows for Scalable and Regularized Barycenter Computation" was accepted at @UncertaintyInAI ! In this paper, we provide a gradient flow view of empirical Wasserstein barycenters.

[Eduardo Fernandes Montesuma @efernand9]

This is joint work w/ @YBendou and @mikegartrell , on my research line about Wasserstein barycenters and their applications to Machine Learning.

[Eduardo Fernandes Montesuma @efernand9]

Our paper "ReBaPL: Repulsive Bayesian Prompt Learning" has been accepted at #CVPR2026 as a conference paper. We show how Bayesian MCMC methods explores the multi-modality of the posterior distribution in Prompt Learning

Preprint:
https://t.co/8eetrPcRRM

[Frank Nielsen @FrnkNlsn]

🎓Get the latest results in information geometry! 😊

A collage of about 150 papers published in the first issues of the "Information Geometry" journal (Springer INGE):

https://t.co/sn8r7sMwSF

@SN_INGE

˙✧˖°🎓 ༘⋆｡ °

[Eduardo Fernandes Montesuma @efernand9]

@ViktorStein2 Thanks! Me too :). Send me a DM during the conference!

[Eduardo Fernandes Montesuma @efernand9]

My paper on Knowledge Distillation has been accepted at the 7th International Conference on Geometric Science of Information (GSI'25). In this paper I explore Feature Knowledge Distillation from a theoretical point of view, and how it performs under different probability metrics

[Stat.ML Papers @StatMLPapers]

KD$^{2}$M: An unifying framework for feature knowledge distillation https://t.co/KYEq92d2So

[Eduardo Fernandes Montesuma @efernand9]

I started posting a series of toy examples in Optimal Transport for Machine Learning, mostly related to my survey "Recent advances in optimal transport for machine learning" and my PhD thesis:

https://t.co/KNAZ63J7mY

[Eduardo Fernandes Montesuma @efernand9]

For more details, discussions and definitions on the used metrics, check out the main paper!

[Eduardo Fernandes Montesuma @efernand9]

We tested different probability metrics for the KD task, based on Empirical and Gaussian approximations. We also considered supervised variants of these metrics, showing that these actually perform (slightly) better than unsupervised variants.