Alejandro García Castellanos

Alejandro García Castellanos

PhD Candidate

AMLab @ UvA

About me

I am a PhD candidate at University of Amsterdam, focusing on applying topology, algebra, and geometry in machine learning.

Interests
  • Representation Learning
  • Geometric Deep Learning
  • Topological Machine Learning
  • Non-Euclidean Geometry
Education
  • MSc in Machine Learning, 2023

    KTH Royal Institute of Technology

  • BSc in Mathematics and Computing, 2021

    Universidad Politecnica de Madrid

Publication List

(2024). HyperSteiner: Computing Heuristic Hyperbolic Steiner Minimal Trees. arXiv preprint arXiv:2409.05671.

PDF Cite Code

(2024). Learning symmetries via weight-sharing with doubly stochastic tensors. ICML 2024 Workshop on Geometry-grounded Representation Learning and Generative Modeling.

PDF Cite

(2024). Relative Representations: Topological and Geometric Perspectives. arXiv preprint arXiv:2409.10967.

PDF Cite Code

(2023). Topological regularization and relative latent representations. KTH Royal Institute of Technology.

PDF Cite Code Slides

Timeline

 
 
 
 
 
PhD Candidate
AMLab @ UvA
February 2024 – Present Amsterdam, Netherlands
Under the supervision of Erik Bekkers (University of Amsterdam) and co-supervision of Daniël Pelt (University of Leiden), we will develop techniques for collaborative human-computer image annotation of training sets for deep learning tasks. These techniques will suggest relevant annotations to the human annotator, will indicate inconsistencies in the human annotations, and will use concepts from geometric deep learning to handle shapes of image annotations.
 
 
 
 
 
Research Engineer
Division of Robotics, Perception and Learning @ KTH
March 2023 – February 2024 Stockholm, Sweden

Projects under the supervision of Danica Kragic:

  • I delved into Geometric Deep Learning and Lie groups while working on a project that involved devising path-finding algorithms on learned equivariant representations through class-pose decomposition.
  • I also explored Manifold Learning techniques, employing probabilistic models to understand data shapes, following the methodology of Georgios Arvanitidis (2021).
  • Lastly, I worked on a project that proposes a method to recover the underlying hierarchies within hyperbolic embeddings.
 
 
 
 
 
Trainee Internship
Medical Data Analytics Laboratory @ UPM
September 2020 – January 2021 Madrid, Spain

Projects under the supervision of Ernestina Menasalvas Ruiz:

  • Application of information recognition algorithms in clinical notes.
  • Development of new techniques for the recognition of metrics, doses and numbers.
  • Fine-tuning a BERT model in breast and lung cancer’s clinical notes.