Alejandro Calle-Saldarriaga

Alejandro Calle-Saldarriaga

5th year Statistics PhD candidate working at the interface of Bayesian modeling and deep learning for large-scale scientific data: scalable Gaussian processes, deep generative models, and uncertainty quantification, mostly for climate and atmospheric problems. Graduating in 2027 and looking for jobs.

Sitemap

A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.

Pages

Posts

publications

Amortized probabilistic retrieval of atmospheric CO2 from OCO-2 spectra using deep learning with Laplace approximations and normalizing flows

Published in Preprint (under review), 2026

An amortized neural inference framework for calibrated, non-Gaussian uncertainty in satellite CO2 retrieval. Preprint, under review.

Recommended citation: Calle-Saldarriaga, A.; Jimenez, F.; Grosskreuz, J.; Wang, J.; Hobbs, J.; Katzfuss, M. (2026). "Amortized probabilistic retrieval of atmospheric CO2 from OCO-2 spectra using deep learning with Laplace approximations and normalizing flows." Preprint, under review. https://arxiv.org/abs/2606.17413

Generative multi-scale modeling via spatial autoregressive transport maps

Published in Technometrics, 2026

A multi-scale Bayesian transport map model for generative modeling of nonstationary spatial fields.

Recommended citation: Calle-Saldarriaga, A.; Wiemann, P.F.V.; Katzfuss, M. (2026). "Generative multi-scale modeling via spatial autoregressive transport maps." Technometrics.

talks

A non-learning premise selection algorithm for Apia

Published:

Abstract: Automatic theorem provers need to recieve a reasonably small number of premises in order for them to be able to prove a given conjecture with limited processor time. In large theories this is not always possible, as many irrelevant clauses are added to the premises. In order to solve this problem, premise selection algorithms have emerged in the past few years, some using non-learning methods and others using learning ones. Our goal in this project is to implement a non-learning premise selection algorithm for Apia, in order to further link the interactive theorem prover Agda with Authomatic theorem provers. Slides

A set theory formalization

Published:

Abstract: Set theory has been is one of the most important fields for the foundations of mathematics, so having a formalization (i.e. a translation of its axiom and theorems into some proof assistant) is desirable. In this talk, we present a set theory formalization of the Z axioms and some theorems using Agda. Slides

A homogeneity test based on depth-depth plots for functional data

Published:

Abstract: One of the classic concerns in statistics is determining if two samples come from the same population, i.e., homogeneity testing or two-sample testing. In this paper, we propose a homogeneity test in the context of Functional Data Analysis, adopting an idea from multivariate data analysis: the data depth plot (DD-plot). This DD-plot is a generalization of the univariate Q-Q plot (quantile-quantile plot). We propose some statistics based on these DD-plots, and we use bootstrapping techniques to estimate their distributions. We simulate our test’s finite-sample size and power, obtaining better results than other homogeneity tests proposed in the literature. Finally, we illustrate the procedure in samples of real heterogeneous data and get consistent results.

Generative multi-scale modeling via spatial autoregressive transport maps

Published:

Talk on multi-scale Bayesian transport maps, a scalable Bayesian emulator for nonstationary spatial fields that models multiple resolutions jointly through autoregressive transport maps, applied to climate model downscaling with full posterior uncertainty quantification.

teaching

CEC

Continuous Education, Universidad EAFIT, Departmento de Ciencias Matemáticas, 2022

Led continuous education courses, mostly related to basic statistics or machine learning.

Datos Funcionales - CM0867 - 2022-1

Master, Universidad EAFIT, Departmento de Ciencias Matemáticas, 2022

Functional data analysis course for Master Students at EAFIT. Topics covered: exploratory fda, hilbert spaces, smoothing, fpca, functional linear model. Lecture notes here

UW Madison TA

TA, UW Madison, 2022

Teaching assistant for Stats 301 (Introduction to statistical methods), Stats 311 (Introduction to theory and methods of mathematical statistics), Stats 312 (Introduction to theory and methods of mathematical statistics II), Stats 371 (Introductory applied statistics for the life sciences), and Stats 461 (Financial statistics).