The girl in the picture up there is me.
A couple of unrequested facts about me. I like coffee, sunny seaside towns, honesty and British humor. In the ever-lasting fight between cats and dogs I side with dogs (but sweet cats can easily win me over). My spirit is mid-way between Hugh Grant and Julia Roberts. Basically I’m a Notting Hill kid. I have many addictions, but they never stay the same. When I’m not working, I’m watching movies, doing sports or asking very odd questions (e.g. when was the concept of breakfast born?)
I believe in hard work and commitment, but I value a balanced and healthy perspective on life above any form of excellence. My opinions are always evolving, so don’t take anything you read here as absolute truth—if there’s any such thing.
This is my blog, which is a space where I feel free to share my thoughts and most random interests. Just like in my life, there’s no fixed theme or grand vision. If you like some of my thoughts, or equally if you don’t and have something to tell me feel free to send me an email. That’s all for now.
Research bio
I do research on continual learning. If you want to know why, check out this page.
I'm a last-year doctoral fellow at the ETH AI Center, advised by Thomas Hofmann and Benjamin Grewe. I use deep learning theory (learning dynamics and optimization) to build foundations for continual learning algorithms. Much of this is joint work with Razvan Pascanu. During the PhD I spent a few months at RIKEN AIP in Tokyo (Emtiyaz Khan's lab, 2025) and at Mila in Montréal (Doina Precup's group, 2026).
Primitives of Memory: Foundations of Continual Learning Algorithms, Mila Tea Talk, July 2026.
News
Jul 26🎙️ Mila Tea Talk, Primitives of Memory: Foundations of Continual Learning Algorithms. recording
Jul 26🎙️ ELLIS × UniReps Speaker Series, One Principle, Many Methods: Function-Space Invariance in Continual Learning. Session shared with Razvan Pascanu.recording
Jul 26🔗 To Retain or to Adapt? Generalizing Continual Learning is on arXiv. With Mandana Samiei, Doina Precup, Razvan Pascanu, Claire Vernade.paper
Jul 26🔗 Provable Forgetting Bounds Drive Capacity Savings: Spectral Thresholding in Continual LoRA, ICML 2026 workshop Continual Adaptation at Scale. With Louis Barinka, Enis Simsar, Tanguy Dieudonné, Thomas Hofmann.
Jul 26📝 Under review: Continual Learning Algorithms(with Razvan Pascanu) and Resurrecting Regularization in Continual Learning(with Naima Elosegui Borras, Amir Joudaki, Thomas Möllenhoff, Thomas Hofmann, Razvan Pascanu). Preprints soon.
Jun 26🔗 When One Adapter Speaks for Many: Discovering Low-Rank Redundancy in Continual Fine-Tuning, ICML 2026 workshop on Connecting Low-rank Representations in AI. With Tanguy Dieudonné, Enis Simsar, Louis Barinka, Thomas Hofmann.paper
Jun 26🍁 Visiting researcher at Mila, Montréal, in Doina Precup's group, until the end of July.
Apr 26🔗 Barriers for Learning in an Evolving World: Mathematical Understanding of Loss of Plasticity, ICLR 2026, Rio. With Amir Joudaki, Mohammad Sadegh Razlighi, Iman Mirzadeh, Keivan Alizadeh, Thomas Hofmann, Mehrdad Farajtabar, Fartash Faghri.paper
Apr 26🔗 Heads Collapse, Features Stay: Why Replay Needs Big Buffers, ICLR 2026, Rio. With Damiano Meier, Thomas Hofmann.paper · slides
