Pablo Robin Guerrero

I recently completed my Bachelor’s degree in Communication Systems at EPFL, with a strong focus on theoretical computer science and machine learning. My academic interests lie at the intersection of representation learning, computational modeling, and the study of communication and behavior.

I am currently a Research Engineer at École Normale Supérieure in the Neural Circuit dynamics & Behaviour Laboratory. My work focuses on the social and latent acoustic structure of bottlenose dolphin communication, aiming to link whistles to behaviour. In particular, I investigate codebook structure and learn acoustic representations from dolphin vocalizations to identify phoneme-like units and characterize the long-term temporal organization of whistle sequences, and I relate these representations to temporal structure and behavioural context to probe how dolphin vocal sequences are organized.

Previously, I have conducted research on efficient machine learning systems and inference optimization. At KAIST, EPFL and Tsinghua University, I worked on efficient deployment of vision-language models on mobile devices and latent-space routing for generative inference, studying efficiency–performance trade-offs under constrained hardware settings.

More broadly, I am interested in cognition and linguistics, with a focus on how structured representations emerge from complex signals and support communication and behavior. I am particularly interested in computational approaches that use representation learning to study language, vocal communication, and other high-dimensional sensory data.