Computational mathematics · Edinburgh, UK

Alessandro Carraro

PhD student in computational mathematics

I research structure-preserving numerical linear algebra methods to make deep learning models more compact, efficient, and scalable.

Download CV GitHub LinkedIn

University of Edinburgh
September 2026 – present

What I work on

Research

I am a PhD student at the University of Edinburgh, where I study AI model compression through structured linear algebra. I completed the joint Computational Science and Engineering master’s programme at KTH and TU Delft, graduating cum laude.

My research explores questions such as: Can we exploit tensor decompositions or structured operators to reduce the memory footprint of neural networks? What are the trade-offs between memory usage, computational cost, and model accuracy? And how can we translate these mathematical ideas into efficient GPU implementations?

Structured linear algebra Machine learning CUDA Scientific software

Education & experience

Sep 2026 – present

PhD in Computational Mathematics

University of Edinburgh

May – Aug 2026

Google Summer of Code contributor

NextLA.jl

Designed and implemented GPU-accelerated Tile-Low-Rank matrix multiplication, achieving up to 2.8× speedup over the existing implementation.

Sep 2024 – June 2026

MSc in Applied Mathematics

KTH Royal Institute of Technology & Delft University of Technology

Joint degree, cum laude. Thesis: A Mixed-Precision Block-Jacobi SVD for Modern GPUs.

May – Aug 2025

Google Summer of Code contributor

JuliaReach

Implemented reachability algorithms for the analysis of uncertain dynamical systems in Julia.

Sep 2024 – Mar 2025

Graduate Teaching Assistant

Delft University of Technology

Supported master’s and bachelor’s courses in high-performance computing, including MPI, OpenMP, and CUDA.

Sep 2021 – Jun 2024

BSc in Aerospace Engineering

Delft University of Technology

Minor in Computational Science and Engineering.

Selected work

Projects & Publications

Numerical linear algebra · CUDA C++

Multi-precision Block-Jacobi SVD

Developed a mixed-precision SVD algorithm for modern GPUs. It achieved up to 5× speedup over a full-FP64 implementation while retaining comparable accuracy on ill-conditioned problems.

MSc thesis · Jan – Jun 2026

Read the thesis

GPU computing · Julia

Tile-Low-Rank GEMM in NextLA.jl

Built GPU-enabled Tile-Low-Rank matrix multiplication routines and extended the open-source library’s structured linear algebra tools.

Read the project write-up

Workshop paper · ARCH-COMP 2026

ARCH-COMP26 Category Report: Continuous and Hybrid Systems with Linear Continuous Dynamics

Co-authored the ARCH-COMP26 category report on continuous and hybrid systems with linear continuous dynamics, published in EPiC Series in Computing.

EPiC Series in Computing, Vol. 110 · 2026

Read the workshop paper

Occasional technical notes and project logs: browse the archive