This repository holds the reproducible sources of the PhD thesis:
de Souza, L. (2024). Physics-informed machine learning: application to Seismic Imaging [Doctoral dissertation, Université Paris sciences et lettres]. https://doi.org/10.70675/c88f4eaaz38e6z4704za3ccz9350a54a8f02
The thesis is archived and indexed at:
- HAL open archive: https://pastel.hal.science/tel-05634155
- theses.fr: https://theses.fr/2024UPSLM077
- Sudoc: https://www.sudoc.fr/297551876
The latest version of the thesis can be downloaded from https://github.com/ldeso/phd-thesis/releases.
Download the sources by cloning this git repository.
git clone https://github.com/ldeso/phd-thesis.git
cd phd-thesis
Each version of the thesis is associated with a unique tag. For example, the version that was validated by the PhD jury is tagged v1.0.0.
git checkout v1.0.0
The thesis can be built from the sources in three steps.
Step 1: install the TeX Live distribution from 2025.
curl -O https://texlive.info/historic/systems/texlive/2025/tlnet-final/install-tl-unx.tar.gz
tar xzf install-tl-unx.tar.gz
./install-tl-20260301/install-tl -portable -texdir texlive -scheme scheme-small \
-repository https://texlive.info/historic/systems/texlive/2025/tlnet-final -no-interaction
rm -r install-tl-unx.tar.gz install-tl-20260301
Step 2: install the required LaTeX packages.
./texlive/bin/x86_64-linux/tlmgr install biber biblatex boondox cochineal \
courier csquotes eepic endfloat enumitem fontaxes helvetic imakeidx latexmk \
logreq ly1 multirow newtx overpic pmboxdraw refstyle relsize scrhack siunitx \
soul sourcecodepro tagpdf txfonts varwidth xstring
Step 3: build the file phd-thesis.pdf.
PATH="$PWD/texlive/bin/x86_64-linux:$PATH" ./texlive/bin/x86_64-linux/latexmk \
-cd -gg -pdf src/phd-thesis.tex
Chapter 5 "Bayesian physics-informed neural networks with explicit loss weights" is also available as a standalone article preprint. It can be built on its own from the sources in src/chapters/pinn-spde/.
PATH="$PWD/texlive/bin/x86_64-linux:$PATH" ./texlive/bin/x86_64-linux/latexmk \
-cd -gg -pdf src/chapters/pinn-spde/preprint.tex
The simplest way to edit the sources is by using the LyX editor.
PATH="$PWD/texlive/bin/x86_64-linux:$PATH" lyx src/phd-thesis.lyx
Make sure to reconfigure LyX at least once so that it is made aware of all the LaTeX packages installed above ("Tools ▸ Reconfigure").
After the edits, export the sources to the LaTeX format using the LyX interface ("File ▸ Export ▸ LaTeX (pdflatex)") or directly from the command line.
lyx -batch -f all -e pdflatex src/phd-thesis.lyx
Chapter 5 can be edited and exported as a standalone LaTeX file in the same way.
lyx -batch -f all -E pdflatex src/chapters/pinn-spde/preprint.tex \
src/chapters/pinn-spde/pinn-spde-article.lyx
If you use or refer to the thesis, please cite it as follows.
@phdthesis{de_Souza_Physics-informed_machine_learning_2024,
title = {{Physics-informed machine learning : application to Seismic Imaging}},
author = {de Souza, Léo},
url = {https://pastel.hal.science/tel-05634155},
school = {Université Paris sciences et lettres},
year = {2024}
month = nov,
doi = {10.70675/c88f4eaaz38e6z4704za3ccz9350a54a8f02},
}
If you specifically use this repository, cite it as described in CITATION.cff.
This repository follows the REUSE specification: every file's license is declared in its header or in REUSE.toml, with the license texts in LICENSES/. In summary: the thesis text, figures, and sources are licensed under CC BY 4.0. Accompanying code is dual-licensed CC BY 4.0 or MIT, except vendored LaTeX files that keep their own licenses, and the institutional cover artwork that is reproduced with permission (see LicenseRef-MinesParisPSL.txt).