01 / IntroductionA Coruña, Spain
NeurIPS 2026Space-Aware World Models

Bruno Corcuera.

I work on world models, focusing on representation learning and generative modeling.

Doctoral Researcher at LIDIA, Universidade da Coruña, supervised by Brais Cancela.

Publications

PreprintarXiv · 2025

Unsupervised Dynamic Feature Selection for Robust Latent Spaces in Vision Tasks

Bruno Corcuera, Carlos Eiras-Franco, Brais Cancela

Learning which visual features to keep: unsupervised, instance-specific selection for more robust latent representations.

Background

Oct 2026 –Present

Doctoral Researcher

LIDIA · Universidade da Coruña
PhD in Machine Learning · Supervisor: Brais Cancela

Research on world models and efficient generative models.
Apr 2025 –Sep 2026

Research Assistant

LIDIA · Universidade da Coruña

Research on representation learning and dynamic feature selection for efficient visual deep learning models.
2025–26

MSc in Business Intelligence & Big Data in Secure Environments

Universities of Burgos, León & Valladolid

Master’s thesis on a local generative-AI assistant for pedagogical code review.
Sep 2024 –Feb 2025

Data Scientist Intern

Inditex

Worked on large-scale sales forecasting and production ML pipelines. Built training and deployment workflows using Snowflake, Azure ML, Azure DevOps, Spark and Databricks, including distributed training.
2021–25

BSc in Data Science & Engineering

Universidade da Coruña

Thesis on perceptual latent representations for world models. Awarded honours and the AEPIA Best Bachelor’s Thesis Award, 1st prize (2026).

Projects & code

Selected research projects and implementations, with their code.

Space-Aware World Models (SWM)

NeurIPS 2026 · world models · spatial representations

Factored scene representations for spatially coherent rollouts and novel environment synthesis.

Perceptual Latent Representations for World Models

Research project · representation learning · world models

Learning compact perceptual representations for improved reconstruction, learned dynamics and downstream control.

AEPIA best thesis award.

World Models in PyTorch

Implementation · world models

End-to-end PyTorch implementation of World Models with VAE vision, MDN-RNN dynamics and CMA-ES control.

3D Gaussian Splatting under a Memory Budget (Budget-3DGS)

Research project · neural rendering · novel view synthesis

Splitting Gaussians where the render still misses detail, for sharper scenes from the same memory as 3DGS-MCMC.

Fourier Feature Encodings vs. MLP Depth

Research project · spectral bias · implicit representations

Studying how Fourier input encodings substitute for MLP depth across signals, images, 3D shapes and tabular data.

Latent Market Volatility

Research project · representation learning · financial time series

Probing how a small transformer’s latent space captures stock co-movement, sector structure and market stress while forecasting volatility.

All repositories

Contact

bruno.sanchez1@udc.es

Personal email: brunocorcueras@gmail.com