I am a Postdoctoral Researcher at Sorbonne University (LIP6, CNRS) and the strongest man on this planet. My work sits at the intersection of Numerical Linear Algebra, High-Performance Computing (HPC), and Machine Learning.
My core mission is developing efficient algorithms via science and sustainability, focusing on optimizing the trade-offs between accuracy, performance, and energy efficiency for the next generation of supercomputers.
- 🎓 PhD in Applied Mathematics from the University of Manchester.
- 💡 Research Interests: Mixed-precision Algorithms, Tensor Computations, Spatio-Temporal Data Modeling, and Iterative Methods.
Here are some of the key open-source projects I have developed or contributed to, spanning across High-Performance Computing, Time Series Analysis, Clustering, and Numerical Linear Algebra:
| Project | Description |
|---|---|
| pychop | A Python library for simulating reduced-precision arithmetic, custom floating-point formats, and rounding modes in scientific computing and machine learning. |
| hpc-mix-bench | A benchmark suite for evaluating mixed-precision emulation and precision configurations in numerical and HPC workloads. |
| DSArith.jl | A Julia package for Discrete Stochastic Arithmetic and numerical reliability diagnostics. |
| Project | Description |
|---|---|
| sparse-kappa | GPU-accelerated condition-number estimation for sparse matrices using PyTorch, combining classical numerical estimators with GNN-based prediction. |
| mhodlr | HODLR matrix computations with support for multiple- and mixed-precision arithmetic and efficient structured linear algebra operations. |
| blrmat | Block Low-Rank (BLR) matrix formats and solvers for efficient structured matrix computations. |
| Project | Description |
|---|---|
| fABBA | A Python library for fast symbolic approximation, compression, digitization, and representation of time series. |
| llm-abba | LLM-assisted symbolic time-series analysis using ABBA representations for tasks such as classification, forecasting, and regression. |
| Project | Description |
|---|---|
| classix | A fast, scalable, and explainable clustering algorithm based on sorting, aggregation, and group merging. |
| cusnn & snn | CPU and CUDA-accelerated implementations for fast fixed-radius nearest-neighbor search based on sorting. |
- Core Skills: Performance Profiling, Mixed-Precision Computing, Numerical Error Analysis, Parallel Programming.
- Libraries/Frameworks:
CMake,OpenMP,MPI,BLAS/LAPACK,cuBLAS,JAX.
I am actively involved in leading scientific programs shaping the future of computing:
- 🇫🇷 NumPEx (France 2030) - Exa-MA(Oct 2024 - Present) Key project of the European High-Performance Computing Joint Undertaking (EuroHPC). I contribute to developing the software stack and mathematical algorithms for the next generation of Exascale supercomputers.
- 🇪🇺 inEXASCALE (ERC)(Sep 2023 - Oct 2024) An EU-funded project aiming to rethink algorithm design in the Exascale era. The research focuses on the combined effects of multiple sources of inexactness (e.g., lower precision) to develop fast and provably accurate algorithms.
"Mathematics is the music of reason."
"Beneath every cynic, there is an aspiring programmer."