Pablo Miralles

Hi, I’m Pablo! I’m a PhD student in Artificial Intelligence supervised by Alejandro Martín and Javier Huertas-Tato. Before this, I completed an MSc in Machine Learning and Big Data at the Technical University of Madrid, and a double BSc in Mathematics and Computer Science at the University of Murcia.

My research focuses on applying deep learning to natural language processing (NLP), with work spanning AI-generated text detection, authorship attribution, natural language inference (NLI), and efficient Transformer architectures. Currently, I’m interested in how large language models (LLMs) can be adapted to downstream tasks in a way that generalizes robustly, especially when data are scarce or affected by spurious correlations. I primarily work with PyTorch, HuggingFace, and vLLM. I’m driven by understanding the mathematical and intuitive foundations of models, the structure of the data, and the demands of the pre-training and downstream tasks, using that insight to design methods that are more efficient, robust, and interpretable in real-world settings.

Feel free to check out my publications, occasional posts, or reach out if you’re interested in collaborating!