Machine learning research · Computer vision · Reliable systems
I'm Nassim Arifette. My work spans 3D medical imaging, computer vision, neural-network verification, and structured learning. I build reproducible systems and evaluate them carefully, from anatomy-preserving CT-to-MRI translation to constraint-aware models and grounded AI assistants.
Select a question
Selected question
Can an unpaired 3D model change imaging modality while preserving anatomy and voxel geometry?
Fully volumetric CT-to-UTE-MRI translation, controlled CycleGAN comparisons, and structure-aware losses.
Selected theme: Medical image translation. Can an unpaired 3D model change imaging modality while preserving anatomy and voxel geometry?
Research threads
These are recurring themes across the work already documented here. Each one connects a concrete question to a project, report, or experiment.
01
Can an unpaired 3D model change imaging modality while preserving anatomy and voxel geometry?
Fully volumetric CT-to-UTE-MRI translation, controlled CycleGAN comparisons, and structure-aware losses.
02
How can geometric and safety constraints be carried through learned models?
Set-based ReLU verification and metric projection for constraint-aware phylogenetic models.
03
Which representations retain the structure a task needs when data or model capacity is limited?
Comparative work on speech features and knowledge-graph embedding families under controlled evaluation.
04
How can retrieval and tool use turn open requests into useful plans tied to data and context?
A multilingual RAG and tool-using concierge that turns visitor questions into navigable plans.
Selected results and artifacts
Unpaired 3D medical image translation with custom losses.
Overview During my internship with the MRI group at BioMaps, I studied unpaired, fully volumetric CT to UTE MRI translation of the thorax . The practical goal was label transfer: generate MRI like volumes from annotated CT scans while keepi...
An exploratory study of abstraction-based ReLU verification using zones (DBMs), tropical geometry, and Julia prototypes.
Verification Neural Network Institution: École Polytechnique, LIX Degree: Master 1 research project <br / Date: March 26, 2024 Author: Nassim Arifette Supervisors: Éric Goubault, Sylvie Putot Abstract This thesis addresses a critical challe...
Comparative analysis of deep learning approaches for entity alignment in knowledge graphs (YAGO/Wikidata/DBpedia) at LISN, Univ. Paris-Saclay.
Institution : LISN – Laboratory for Interdisciplinary Numerical Sciences, University of Paris Saclay Degree : Master’s Internship Dates : June–July 2024 Author : Nassim Arifette Supervisors : Dr. Fatiha Saïs, Olivier Inizan Project Overview...
A multilingual AI concierge that turns open questions into concrete day plans across the Versailles estate.
“Where should I start? Will it rain? Can I still see the fountains?” During the Datacraft × Château de Versailles hackathon, we kept hearing these questions from visitors and staff alike. Versailles Concierge is the assistant we built to an...
Histogram-aware training improved distributional similarity while retaining anatomical fidelity.
Report values comparing the ResNet-LSGAN baseline with Hist-CycleGAN.
Research notebook
Longer explanations, mathematical readings, implementation notes, and working ideas. The aim is to make the path to a result visible, not only the result.
Browse all notesSeriesCategory Theory for AI
An introductory overview of how category theory concepts can be applied to artificial intelligence.
This series delves into the intersection of category theory and artificial intelligence, providing insights into how abstract mathematical concepts can inform and enhance the design of AI systems. Each part builds on the previous one, so st...
SeriesCategory Theory for AIPart 2
A worked path through Seely's 2-categorical lambda calculus: from one term with two beta reductions to interchange, laxity, and the adjunction between currying and uncurrying.
Fix a type $O$, a closed function $f:O\Rightarrow O$, and one free input $y:O$. Consider the term $$ T(y) = (\lambda u:O.\,u) \big((\lambda v:O.\,v)(f\,y)\big). $$ It contains two beta redexes. We can contract the outer one first, $$ T(y) \...
SeriesAI FoundationsPart 4
A story-driven tour of new papers and benchmarks—plus what actually moved the needle in my pipelines, with links and practical takeaways.
Two bugs broke my RAG system last month. The first was self‑inflicted—I nudged top‑k from 8 to 16 “for coverage” and watched answers get longer and less faithful. The second came from a knowledge base that contradicted itself; my agent chee...
Contact