2026

1. May 2026

   arXiv

   MAPS: A Synthetic Dataset for Probing Vision Models in a Controlled 3D Scene Space

   Santiago Galella, Pamela Osuna-Vargas, Maren Wehrheim, Martina G. Vilas, Gemma Roig, and Matthias Kaschube

   A synthetic 3D dataset for probing what vision models represent, with full control over the scene.

   arXiv preprint arXiv:2605.20549

   arXiv Bib

   @article{galella2026maps,
     title = {{MAPS}: A Synthetic Dataset for Probing Vision Models in a Controlled {3D} Scene Space},
     author = {Galella, Santiago and Osuna-Vargas, Pamela and Wehrheim, Maren and Vilas, Martina G. and Roig, Gemma and Kaschube, Matthias},
     journal = {arXiv preprint arXiv:2605.20549},
     year = {2026},
     month = may,
   }

   

2. May 2026

   bioRxiv

   Contributing author

   Random Network Structure Stabilizes Neural Manifolds

   Jens-Bastian Eppler, Santiago Galella, Gabriel C. Mel, and Alex Roxin

   Random connectivity keeps the similarity structure of neural representations stable, even as synapses keep changing.

   bioRxiv

   DOI Bib

   @article{eppler2026random,
     title = {Random Network Structure Stabilizes Neural Manifolds},
     author = {Eppler, Jens-Bastian and Galella, Santiago and Mel, Gabriel C. and Roxin, Alex},
     journal = {bioRxiv},
     year = {2026},
     month = may,
     doi = {10.64898/2026.05.21.726949},
   }

   

2025

1. Dec 2025

   NeurIPS

   Dimensionality Mismatch Between Brains and Artificial Neural Networks

   Santiago Galella, Maren Wehrheim, and Matthias Kaschube

   Representations become higher-dimensional along the human visual hierarchy, but collapse in the later layers of deep vision models, limiting their semantic generalization.

   Advances in Neural Information Processing Systems (NeurIPS)

   Paper Bib

   @inproceedings{galella2025dimensionality,
     title = {Dimensionality Mismatch Between Brains and Artificial Neural Networks},
     author = {Galella, Santiago and Wehrheim, Maren and Kaschube, Matthias},
     booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
     volume = {38},
     pages = {43679--43716},
     year = {2025},
     month = dec,
   }

   

2024

1. 2024

   bioRxiv

   Internal Neural Representations in Task Switching Guided by Context Biases

   Santiago Galella and Salva Ardid

   Context biases shape shared internal representations in networks that switch between tasks.

   bioRxiv

   DOI Bib

   @article{galella2023internal,
     title = {Internal Neural Representations in Task Switching Guided by Context Biases},
     author = {Galella, Santiago and Ardid, Salva},
     journal = {bioRxiv},
     year = {2024},
     doi = {10.1101/2023.07.24.550365},
   }

   

2020

1. Jun 2020

   Chaos

   Soft-Wired Long-Term Memory in a Natural Recurrent Neuronal Network

   Miguel A. Casal^*, Santiago Galella^*, Oscar Vilarroya, and Jordi Garcia-Ojalvo * Equal contribution

   The recurrent neural network of C. elegans can support long-term memory without changes in its wiring.

   Chaos: An Interdisciplinary Journal of Nonlinear Science

   DOI Bib

   @article{casal2020softwired,
     title = {Soft-Wired Long-Term Memory in a Natural Recurrent Neuronal Network},
     author = {Casal, Miguel A. and Galella, Santiago and Vilarroya, Oscar and Garcia-Ojalvo, Jordi},
     journal = {Chaos: An Interdisciplinary Journal of Nonlinear Science},
     volume = {30},
     number = {6},
     pages = {061101},
     year = {2020},
     month = jun,
     doi = {10.1063/5.0009709},
   }