Alberto Maria Pepe

PhD, University of Cambridge

Project Figure

PhD candidate at the University of Cambridge, supervised by Joan Lasenby and advised by José Miguel Hernández-Lobato. Working on AI Agents, VLMs, Computer Vision and Hypercomplex Neural Networks. Fascinated by how machines see, understand and process geometrical objects and spatial transformations. Previously at Microsoft ('22, '24, '25) and Huawei ('23).

scholar researchgate linkedin github mail
Experience
Microsoft - Remote - May 25 - Jul 25
Vision-Language Models, Applied Sciences Group.
Microsoft - Redmond, WA - Jun 24 - Sep 24
Diffusion Models for Computer Vision, Applied Sciences Group.
Huawei - Munich, Germany - Oct 23 - Mar 24
Computer Vision (Neural Physics Emulators), Intelligent Cloud Technologies Lab.
Microsoft Research - Cambridge, UK - Feb 22 - May 22
Machine Learning (ML Encoder/Decoder for Optical Links), Cloud Infrastructure Group.
Projects
Project Figure
Geometric Inductive Priors in Diffusion-Based Optical Flow Estimation
ICCV 2025 - BEW, Honolulu, HI
paper / code
Project Figure
Define, Refine, Align: Correspondence-free 3D Line Alignment with Attentional, Equivariant and Rotational Layers
CVPR 2025 - PBVS, Nashville, TN
paper / code / poster / slides
Project Figure
Reinforce Loss for QwenChat LLM [sandbox]
Apr 2025, just for fun
code
Project Figure
Vision Transformer for Weather Forecasting [sandbox]
Mar 2025, just for fun
code
Project Figure
Fengbo: a Clifford Neural Operator pipeline for 3D PDEs in Computational Fluid Dynamics
ICLR 2025, Singapore [with Huawei]
paper / code / poster / slides / video
Project Figure
Conditional Diffusion U-Net in 3D Space for Protein Coordinates Estimation
Feb 2025, just for fun
code
Project Figure
Torch-GA: Building Geometric Algebra Networks in PyTorch
Jan 2025
code
Project Figure
STAResNet: A Network in Spacetime Algebra to Solve Maxwell's PDEs
AGACSE 2024, Amsterdam
paper / code / slides
Project Figure
DPGNN: Differentiable Physics-and Geometry-Assisted Network for 2D Flow Estimation
AGACSE 2024, Amsterdam
abstract+poster
Project Figure
GA-ReLU: an activation function for Geometric Algebra Networks applied to 2D Navier-Stokes PDEs
ICLR 2024, AI4DifferentialEquations in Science, Vienna
paper / code / poster
Project Figure
CGAPoseNet+GCAN: A Geometric Clifford Algebra Network for Geometry-aware Camera Pose Regression
WACV 2024, Waikoloa, HI
paper / code / poster / slides / video
Project Figure
Clifford Group Equivariant Neural Network Layers for Protein Structure Prediction
NLDL 2024, Tromsø
paper / code / poster / slides
Project Figure
CGA-PoseNet: Camera Pose Regression via a 1D-Up Approach to Conformal Geometric Algebra
ArXiv, 2023
paper / code
Project Figure
Modeling orientational features via geometric algebra for 3D protein coordinates prediction
Mathematical Methods in the Applied Sciences, 2023
paper
Project Figure
POSEIDONIA
MIT Media Lab, Ancient Future Technology, 2022
website
Project Figure
Geometric Algebra Models of Proteins for Three-Dimensional Structure Prediction: A Detailed Analysis
Advanced Computational Applications of Geometric Algebra, Springer Nature, 2022
paper
Project Figure
Using a Graph Transformer Network to Predict 3D Coordinates of Proteins via Geometric Algebra Modelling
CGI 2022, ENGAGE Workshop, Geneva / Lecture Notes in Computer Science
paper
Project Figure
Geometric Algebra Models of Proteins for Three-Dimensional Structure Prediction
ICACGA 2022, Denver, CO
paper+slides
Project Figure
Learning Rotations
Mathematical Methods in the Applied Sciences, 2023 / AGACSE 2021, Brno
paper / code
Project Figure
Machine Learning Approaches to Digital Signal Processing for Optical and Wireless Communications
TBSI, 2020
thesis
Talks
Project Figure
Learning with an edge: a practical approach to blades in Geometric Algebra
Microsoft ASG, May 2025
slides
Project Figure
Seeing Through PDEs: an Interpretable Neural Operator Pipeline for Joint Estimation of Physical Quantities
PhysicsX, Invited Talk, Apr 2025
slides
Project Figure
Solving PDEs with Geometric Algebra Networks
Cambridge-Brno Workshop, Apr 2025
slides
Project Figure
Molecular Geometry Optimization through Rotor-based Evolutionary Algorithm
ICCA 2023, Holon, Israel [joint work with Scuola Normale Superiore, Pisa, Italy]
slides
Project Figure
Geometric Algebra in Protein Structure Prediction
Institute of Mathematics, Brno University of Technology, Dec 2021
slides
Education
University of Cambridge - PhD, Oct 20 - May 25
Probabilistic Systems, Information, and Inference Group (PSI²)
Thesis: Machine Learning with Geometric Algebra: Multivectors for Modelling, Understanding and Computing
Examiners: Petar Veličković , Johannes Brandstetter
Tsinghua - UC Berkeley Shenzhen Institute (TBSI) - M.Sc. 18-20
Data Science & Information Technology
Thesis: Machine Learning Approaches to Digital Signal Processing in Optical and Wireless Communication
Final Grade: A, GPA: 3.91/4
Politecnico di Torino - B.Sc. 15-18
Electronic & Communications Engineering
Final Grade: 110/110 cum laude, GPA: 29.6/30
Tongji University - B.Sc. 16-17 (Exchange)
Information Technology Engineering
GPA: 29.2/30