Please RT - Open PhD position in my group at the Donders Center for Neuroscience, Radboud University. We're looking for a PhD candidate interested in developing theories of learning in neural networks. Applications are open until October 20th. Info:

Just got back from a great summer school at Sapienza University https://t.co/2pfre6dYls where I gave a short course on Dynamics and Learning in RNNs. I compiled a (very biased) list of recommended readings on the subject, for anyone interested:

Please find this English masters program in Neurophysics. Study the brain, artificial neural networks and complex systems through the lens of mathematical modeling and physics. Radboud University / Donders (Netherlands) is one of Europe’s leading universities in Neuroscience. 1/3

Our paper on the statistical mechanics of transfer learning is now published in PRL. Franz-Parisi meets Kernel Renormalization in this nice collaboration with friends in Bologna (@fgerace_) and Parma (P. Rodondo, @rosalbapacelli). https://t.co/dcWGStHMMH

Biophysics, Stat Mech and Machine Learning will meet in Trento from July 7th to 11th, 2025 in our StatPhys29 Satellite Workshop "Molecular biophysics at the transition state: from statistical mechanics to AI": https://t.co/89xU1dOpAP. Co-organized with @r_potestio lab.

New paper with @ai_ngrosso @VITAGroupUT @sebastiangoldt “On How Iterative Magnitude Pruning Discovers Local Receptive Fields in Fully Connected Neural Networks“ accepted at the conference on parsimony and learning (@CPALconf ) https://t.co/T0Vb5mLhTJ 1/

Our paper on Wang-Landau sampling in neural networks is now published in TMLR. Here's a thread by @r_potestio.

New paper with @leonlufkin and @ermgrant! Why do we see localized receptive fields so often, even in models without sparisity regularization? We present a theory in the minimal setting from @ai_ngrosso and @sebastiangoldt

If you missed it at the #NeurIPS2024 posters! Work led by @LeonLufkin on analytical dynamics of localization in simple neural nets, as seen in real+artificial nets and distilled by @ai_ngrosso @sebastiangoldt Leon is a fantastic collaborator, and is looking for PhD positions!

Ceci n'est pas un tweet: @aingrosso.bsky.social

Giving a talk on Stat Mech of Transfer Learning at 3pm EST today at the Deepmath conference in Philadelphia. Here's a link for the live stream:

Why does #compneuro need new learning methods? ANN models are usually trained with Gradient Descent (GD), which violates biological realities like Dale’s law and log-normal weights. Here we describe a superior learning algorithm for comp neuro: Exponentiated Gradients (EG)! 1/12

Our work on Wang-Landau in neural network learning (a.k.a. Wang-Learnau) is now on arXiv. We use enhanced sampling to explore how the entire density of states of a loss function is affected by data structure. A collaboration with friends in Trento powered by @MeleMargherita_.

This mini-review is based on Hugo Cui's PhD thesis: https://t.co/DmVhuhlBmz . My advice to him was: "Write something you would have loved to have when you started your PhD!" He did an outstanding job introducing the rich methods he developed. Enjoy and share widely!

#RockinAI #Roccella Day 4 invited speakers: Alessandro Laio, Sebastian Goldt, Pietro Rotondo, Alessandro Ingrosso

I'll be leaving ICTP in September to start as Assistant Professor at @DondersInst, @Radboud_Uni in Nijmegen. Students interested in pursuing a PhD at the border of Machine Learning, Neuroscience and Statistical Mechanics, don't hesitate to contact me.

@fgerace_ @RosalbaPacelli This work would not have happened without Federica Gerace and our friends Rosalba and Pietro. Looking forward to be in Roccella Jonica https://t.co/TSdGR6NXSa to talk about our work and eat a proper amount of nduja.

@fgerace_ @RosalbaPacelli Using a previously developed synthetic model for source-target correlations (Correlated Hidden Manifold Model), we study how the similarity between tasks affects transfer and determines whether fine-tuning is helpful.

@fgerace_ @RosalbaPacelli The effectiveness of Transfer Learning in fully connected networks with one hidden layer can be analyzed in a Bayesian setting, varying the interaction strength between the source and target network.

@fgerace_ @RosalbaPacelli Fine-tuning features extracted from a pre-trained, data-adundant source task can boost generalization on a target task where data is scarce. Combining replicas with kernel renormalization, we built a theoretical framework to analyze Transfer Learning in the proportional limit.