Alessandro Serra

About Me

I am a Ph.D. student in Theoretical Data Science and Scientific Computing at SISSA and at Laboratory of Data Engineering (LADE) at AREA Science Park under the supervision of Alberto Cazzaniga and Alessandro Laio. Broadly, I am interested in the interpretability of transformer models and their applications in AI safety. More specifically, my research focuses on the intersection of causally inspired methods (such as mechanistic interpretability) and geometrical tools applied to unsupervised learning. I collect notes and random toughts in a digital garden I’m currently living in Trieste, a beautiful city on the sea in the north-east of Italy. You can reach me at: aserra[at]sissa.it.

News

• [Dec. 2025] I will present my last paper @ NeurIPS 2025 in San Diego
• [Sept. 2025] Our Paper “The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models” has been accepted at NeurIPS 2025! 🎉
• [May 2025] I am invited speaker at Interpretability in Large Language Models Using Geometrical and Statistical Methods at University of Amsterdam
• [Oct. 2024] I am starting my PhD at SISSA in Theoretical Data Science and Scientific Computing
• [Sept. 2024] Our paper “The Representation Landscape of Few-Shot Learning and Fine-Tuning in Large Language Models” has been accepted at NeurIPS 2024! 🎉

Publications

Powered by Jekyll and Minimal Light theme.