Stefano Sarao Mannelli

@stefsm.bsky.social

Assistant Professor Chalmers and Göteborg University 🌱 ML + CogSci + StatPhys 🔎 stefsmlab.github.io 🌐

Join us! We are opening many postdoc positions both in London and in Gothenburg! London 🔗https://www.lesswrong.com/posts/GTt33CasvWjxxazJw/hiring-principia-research-fellows 📅 deadline: March 26th Gothenburg 🔗 www.chalmers.se/en/about-cha... 📅 deadline: April 1st

Vacancies

Phone +46-317721000Mail addressChalmers University of Technology412 96 GothenburgE-mail and more contact informationOrganisation number 556479-5598

chalmers.se

Gatsby Computational Neuroscience Unit@gatsbyucl.bsky.social · 5mo ago

📢 Job alert - Deep Learning Theory & AI Safety Applications open for a postdoc fellow (@saxelab.bsky.social lab) to study artificial deep networks using techniques from applied maths & stat physics. ⏰ Deadline: 26 Mar 2026 🤝 In collaboration with @stefsm.bsky.social ℹ️ www.ucl.ac.uk/life-science...

Here at #ICLR2025 and it's going to be a busy week! 1. Today at 10am, Optimal Protocols for Continual Learning via Statistical Physics and Control Theory, Poster #342 2. Tomorrow at 10am, A Theory of Initialisation's Impact on Specialisation Poster, #342 3. Speaking at the SCSLWorkshop on Monday.

I really enjoyed working on this. What started as a reproducibility issue ended up revealing a more complex picture of mean-field dynamics than I originally thought, making me rethink the specialisation phenomenon. Soon at @iclr-conf.bsky.social! 🔗 openreview.net/forum?id=RQz...

A Theory of Initialisation's Impact on Specialisation

Prior work has demonstrated a consistent tendency in neural networks engaged in continual learning tasks, wherein intermediate task similarity results in the highest levels of catastrophic...

openreview.net

Clementine Domine 🍊 @CCN@clementinedomine.bsky.social · last yr.

Our paper, “A Theory of Initialization’s Impact on Specialization,” has been accepted to ICLR 2025! openreview.net/forum?id=RQz... We shows how neural network can build specialized and shared representation depending on initialization, this has consequences in continual learning. (1/8)