Curriculum Vitae

Wassim Kabalan — PhD student in computational cosmology (APC, CNRS/IN2P3)

Wassim Kabalan

PhD Candidate in Computational Cosmology

Computational cosmologist building differentiable, distributed cosmological simulations for full-field inference with focus on open-source software development and high-performance computing at scale.

Contact

---

PDF version

Education

Nov 2023 – Dec 2026 (expected)	PhD, Physics of the Universe, Université Paris Cité, Paris, France APC, CNRS/IN2P3 Thesis: Automatically differentiable and distributed Probabilistic Programming for wEAk gravitational LensING inference Advisors: Eric Aubourg, Josquin Errard, Alexandre Boucaud, François Lanusse
Apr 2023 – May 2023	Advanced AI for Data Analysis, École Polytechnique (Executive Education)
Sep 2016 – Sep 2018	M2, Electronics, Electrical & Automation, Université Gustave Eiffel
Sep 2013 – Nov 2016	Licence, Engineering Science, Université Paris-Est Créteil (UPEC)

---

Research & Professional Experience

Dec 2023 – Present	PhD Researcher in Cosmology, APC, CNRS/IN2P3, Paris, France Thesis goal: Develop differentiable, distributed N-body simulations as weak-lensing forward models to enable gradient-based cosmological inference at LSST scale. Implement a CMB component-separation technique with spatially varying foreground spectral parameters to reduce bias in the tensor-to-scalar ratio (r) for full-sky missions (e.g., LiteBIRD). Skills: JAX, NCCL, C++, CUDA, HPC, automatic differentiation, Bayesian inference, NumPyro, CMB, weak lensing
Oct 2019 – Oct 2023	Software Infrastructure Engineer, Dassault Systèmes, Vélizy-Villacoublay, France Optimized CATIA cache/conversion pipeline (C++/Linux); profiling-driven algorithmic changes improved throughput and robustness for large CAD data flows. Led Linux convergence for a large C++ rich client. Built GitLab CI/CD for multi-team releases; automated testing, packaging, and deployments. Skills: C++, performance profiling, cache & I/O optimization, Linux, CI/CD, test-driven development
Jan 2019 – Oct 2019	Data Acquisition Engineer, SERMA (for Renault), Guyancourt, France Wrote a C library to decode binary sensor data from automotive ECUs. Built a parallel post-processing pipeline in Python (multiprocessing) on an 88-core server for high-throughput log processing. Designed automated data-conversion workflows for CAN-bus telemetry and prototype vehicle sensor streams; packaged with CMake. Skills: C, Python, multiprocessing, data pipelines, CAN bus, CMake

---

Talks & Tutorials

• JAXPM: A JAX-Based Framework for Scalable and Differentiable Particle Mesh Simulations • Bayesian Deep Learning Workshop • May 20–23, 2025 • slides

• Bayesian Inference for Cosmology with JAX [Tutorial] • Bayesian Deep Learning Workshop • May 20–23, 2025 • slides

• Massively Parallel Computing in Cosmology with JAX [Tutorial] • CoPhy 2024 • Nov 18–20, 2024 • slides

• Differentiable and distributed Particle-Mesh n-body simulations • LSST FR 2024 • Jun 10–12, 2024 • slides

---

Open-Source Software

• jaxDecomp — Author. JAX based code for multi-GPU 3D domain decomposition and distributed FFTs (NCCL); scales cosmology workloads on GPU clusters.
• JaxPM — Main contributor/Maintainer. Differentiable particle-mesh simulations in JAX; multi-accelerator support for scalable forward modeling and gradient-based inference.
• FURAX — Main contributor/Maintainer. JAX building blocks for inverse problems; used for CMB component separation at Simons Observatory.
• jax-healpy — Main contributor/Maintainer. JAX-native HEALPix utilities for CMB/spherical data; GPU- and autodiff-ready.
• jax-grid-search — Author. Distributed grid search and gradient-based optimization on JAX/Optax.

Contributions
- S2FFT — Contributor. Differentiable spherical & Wigner transforms (JAX/PyTorch). Contribution: CUDA spherical harmonics to reduce JAX JIT time.

---

Publications & Products

Refereed

• Spagnoletti, A., Boucaud, A., Huertas-Company, M., Kabalan, W., and Biswas, B. 2024. Bayesian Deconvolution of Astronomical Images with Diffusion Models: Quantifying Prior-Driven Features in Reconstructions. arXiv:2411.19158 [astro-ph.IM]. Contribution: Set up JAX-based deconvolution code and ran multi-node simulations on Jean Zay HPC cluster.

• Sommer, K., Kabalan, W., and Brunet, R. 2024. Infrared Radiometric Image Classification and Segmentation of Cloud Structure Using Deep-learning Framework for Ground-based Infrared Thermal Camera Observations. EGUsphere Preprint 2024-101. Contribution: Created and ran JAX-based U-Net model on Jean Zay HPC. Code: github.com/ASKabalan/infrared-cloud-

Software

• Kabalan, W., Lanusse, F., Boucaud, A., and Aubourg, E. 2025. jaxDecomp: JAX Library for 3D Domain Decomposition and Parallel FFTs. Submitted to JOSS.

In Preparation

• Kabalan, W., Lanusse, F., Boucaud, A., and Aubourg, E. 2025. JAXPM: A JAX-Based Framework for Scalable and Differentiable Particle Mesh Simulations.

• Kabalan, W., Rizzieri, A., Sohn, W., Beringue, B., Basyrov, A., Chanial, P., Boucaud, A., and Errard, J. 2025. A novel approach to optimize clustering for parametric map-based component separation for upcoming CMB polarization satellites.

Skills

Programming Python (JAX, NumPyro, BlackJAX), C++, CUDA, C, Bash

HPC & GPU Computing NCCL, MPI, multi-node distributed computing, GPU profiling (Nsight), Slurm, parallel I/O

Statistics & Machine Learning Bayesian inference (MCMC, HMC, NUTS), simulation-based inference, gradient-based optimization

DevOps & Software Engineering GitHub Actions/CI, packaging (PyPI), containers, Linux, CMake, test-driven development

---

Languages

French	Native
English	Professional proficiency (C1)
Arabic	Native
German	Basic (A2)

---