Albert's personal website

Personal Projects

2026-03-23T15:07:13Z

Following are some personal projects I’ve worked on in my spare time.

Diet Optimization (2024-2025)

Link | Repository

A tool for optimizing diet based on data from Open Food Facts, Open Prices, Ciqual and Agribalyse.

Uses DuckDB to store and query the data, and SciPy to solve the optimization problem.

Re-implementation of surface reconstruction from point clouds in JAX (2024)

Repository

Re-implementation in JAX of the method presented in the paper “p-Poisson surface reconstruction in curl-free flow from point clouds” by Yesom Park, Taekyung Lee, Jooyoung Hahn, and Myungjoo Kang.

Raymarching in JAX (2023)

Repository | Colab

Interactive raymarching in JAX. Toy project to learn about raymarching.

Tweet visualization (2022)

Link | Repository

A visualization of how Elon Musks tweets are affecting the markets, for the data visualization course at EPFL.

NORD-10 assembler and emulator (2025)

Repository

An assembler and emulator of a subset of the instruction set of the 16-bit minicomputer NORD-10 produced by Norsk Data in 1973.

Smaller projects

TextTV (2025)

Link | Repository

Minimal web app that shows SVT Text TV, with dark mode support and keyboard shortcuts.

Shows the latest available news from SVT by using the same API as the official app.

Open Prices Map (2025)

Link

A simple web app that shows product price data from Open Prices on a map using Leaflet.js.

Albert Sund Aillet

2026-03-23T15:07:13Z

<a href="./mri.html" style="text-decoration: none;">
    <img src="https://avatars.githubusercontent.com/u/73786209?v=4" alt="Albert Sund Aillet" width="200" />
</a>

Contact: albert [at] sundaillet.com

Location: Geneva, Switzerland

GitHub | LinkedIn

Spoken languages: English (fluent), French (native), Swedish (native)

I am a Research/Software Engineer at CERN, where I develop distributed learning software and contribute to research on reliable machine learning, balancing tradeoffs of performance, privacy, robustness and interpretability.

I completed my undergraduate in Engineering Physics at KTH Royal Institute of Technology. For my Bachelor’s thesis, I worked on cell image classification with convolutional neural networks at KTH and Karolinska Institutet, supervised by Prof. Karl Meinke (link to thesis, mirror on this website).

I obtained a Master’s degree in Machine Learning at KTH with an exchange at EPFL. For my Master’s thesis I studied self-supervised pre-training of attention-based models for 3D medical image segmentation at RaySearch Laboratories, supervised by Dr. Jonas Söderberg and Prof. Mårten Björkman (link to thesis).

During my studies I interned at CERN, Tobii and Ericsson.

I am motivated by both theoretical understanding and practical applications of machine learning.

Skills

Languages: Python, JavaScript, SQL, Shell scripting, C/C++

Frameworks: NumPy, JAX, PyTorch, Matplotlib, plotly, pandas, Flask, d3.js

Misc: Unix, Git, Docker, Podman, LaTeX

Personal Projects

Theses and Selected Publications

Expand

A. S. Aillet, F. Frisk, “Assessing the Impact of Stain Normalization on a Cell Classification Model in Digital Histopathology”, 2021, Available: http://www.diva-portal.org/smash/record.jsf?pid=diva2%3A1632706, mirror
A. S. Aillet and S. Sondén, “[Re] Variational Neural Cellular Automata”, in ML Reproducibility Challenge 2022, Available: https://neurips.cc/virtual/2023/poster/74151, https://github.com/albertaillet/vnca
A. S. Aillet, “Self-supervised pre-training of an attention-based model for 3D medical image segmentation”, 2023, Available: https://www.diva-portal.org/smash/record.jsf?pid=diva2%3A1795309.
A. Protani, L. Giusti, A. S. Aillet, et al., “Federated GNNs for EEG-Based Stroke Assessment”, in UniReps: 2nd Edition of the Workshop on Unifying Representations in Neural Models, 2024, Available: https://neurips.cc/virtual/2024/102633.
D. R. Santos, A. Protani, L. Giusti, A. S. Aillet, P. Brutti, and L. Serio, “Feasibility Analysis of Federated Neural Networks for Explainable Detection of Atrial Fibrillation,” in 2024 IEEE International Conference on E-health Networking, Application & Services (HealthCom), 2024, Available: https://doi.org/10.1109/HealthCom60970.2024.10880809.
A. Protani, L. Giusti, C. Iacovelli, A. S. Aillet, D. R. Santos, G. Reale, A. Zauli, M. Moci, M. Garbuglia, P. Brutti, P. Caliandro and L. Serio, “Towards Explainable Graph Neural Networks for Neurological Evaluation on EEG Signals,” in 2024 IEEE International Conference on E-health Networking, Application & Services (HealthCom), 2024, Available: https://doi.org/10.1109/HealthCom60970.2024.10880717.
A. Protani, M. M. Van De Bosch, L. Giusti, H. Silva, P. Cacace, A. S. Aillet, F. Hummel and L. Serio, “Decoder-Free Supervoxel GNN for Accurate Brain-Tumor Localization in Multi-modal MRI”, in Reconstruction and Imaging Motion Estimation, and Graphs in Biomedical Image Analysis, Lina Felsner et al. (eds.), 2024, Available: https://doi.org/10.1007/978-3-032-06103-4_16.

404

2026-03-23T15:07:13Z

Page not found

Return to the homepage

Demo night presentation

2026-03-23T15:07:13Z

Name: Albert Sund Aillet

Project: Diet Optimization

Links:

I am here to present my Obscure side-project:

“Food as fuel”
Usually when I am buying food I think about what is actually the best for my body and what minimizes my impact on the planet,
Depends on the goals and preferences
But what if these could be defined

The side project I built:

An app that uses:
- Price data of food products
- Nutrition data of food products
- Set of requirements and preferences from the user
In order to proposal an optimal diet, minimizing certain aspect and maximizing other.

It uses the following parts:

The open food facts database for up to date product information
The open prices database for price information
The ciqual and ciqual-calnut for approximate micro and macro nutrient for values that are not present on the food packaging.
I get exchange rates from the European Central Bank to be able to compare all prices.
I get estimated climate impact from AGRIBALYSE.

I had to look a lot into the data and do a few SQL joins.

What is the optimization problem: Linear programming, scipy formulation.

(Sorry for latex messing up, I will fix it later)

$$ \begin{align*} \min_{x} \quad & c^T x \\ \text{such that} \quad & A_{ub} x \leq b_{ub}, \\ %& A_{eq} x = b_{eq}, \\ & l \leq x \leq u, \end{align*} $$

Where in our case:

x is a vector with the quantity of each product
c is the costs (can be price or ecological footprint).
A_ub is a matrix that contains all the contsraints (nutritional requirements).
l = 0 and u = ∞ since we don’t want negative amount of products

This can be solved quickly and efficiently with many out of the box solvers. In my benchmarking, since the problem is not sparse, the revised simplex method was the fastest.

The optimal solution is global.

The following features are available:

sliders to change preferences and in real time get an updated list of optimal products to consume.
a map where you can choose which products to include in the optimization.
an editor to change the objective function to minimize.

Technology used (very simple):

Python Flask backend
DuckDB database since I am doing only reads
Scipy linprog for the optimization
Vanilla JS and d3.js for the graphs.

How I deploy:

I have a Makefile with all the deployment commands
I have a VPS with Hetzner where I the app.

Obscure side project that I thought about during my studies and then thought it would be easy and straightforward to finish in a few evenings.