🙋‍♂️ Hello there, I’m Mario!

“La originalidad no debe buscarse, pues entonces es extravagancia. La originalidad consiste en el retorno al origen; así pues, original es aquello que vuelve a la simplicidad de las primeras soluciones” -Antoni Gaudí.

Scientific divulgation: Youtube ; Medium Blog

👨‍🎓 I studied mathematical engineering with my thesis on multimodal and adversarial recommender systems.

🖥️ I have a master’s degree in computer science, as part of an image retrieval research group.

👉 I’m doing a PhD about AI at UPC, Barcelona, Spain. SOCO Group

Currently, I’m working on efficient - randomized neural networks, reservoir computing and green application in & by AI. My research analyze: learning, reasoning and statistical properties of the modeling of recurrent neural networks and transformers.

👨‍💼 Additionally, I have experience in the industry as a data scientist and generative AI software development.

This figure shows how two randomized artificial neural networks classify handwritten digits. For each class (0-9), we applied an interpretability method, taking a test image and overlaying a relevance map computed from the model’s learned features. We can observe that a typical randomized neural network often concentrates on isolated, noisy pixels, suggesting the model uses spurious cues. In contrast, our proposal concentrates on smooth connected segments that trace the digit shape, indicating more coherent, accurate, and robust decision rules. Results published in DOI: 10.1016/j.eswa.2025.129948 .