Federico D'Agostino retweeted
Long-horizon tasks + sparse rewards = no supervision. So, we need dense supervision methods… but how do we evaluate them?
Meet QVal, our latest release👇
Federico D'Agostino retweeted
🚀New Paper
arxiv.org/abs/2606.28551
Everyone obsesses over VLM architectures & training recipes. But what about the data?
Presenting the latest work in the DataComp-series: a testbed for VLM data curation with 1,000+ controlled experiments and some surprising lessons 👀
🧵👇
Federico D'Agostino retweeted
How well can you describe the feature selectivity of a vision neuron … with words? Interpretability has long borrowed from neuroscience — and maybe it can give back too! 🧵
Federico D'Agostino retweeted
🚨 New paper! arxiv.org/abs/2603.06610
Are you sure your post-trained LLM isn’t forgetting something?
Adapting LLMs is known to cause forgetting.
We usually measure it via general knowledge benchmarks.
But if MMLU doesn’t drop…. are you really fine?
Federico D'Agostino retweeted
How can agents learn in long, open-ended tasks where success is rare and rewards are sparse? 👀
🚨 Enter ∆Belief-RL: we show how to use agent’s own belief updates as a dense reward for turn-level credit assignment.
the result? Surprisingly strong generalization.
(1/8) 🧵⬇️
Federico D'Agostino retweeted
Federico D'Agostino retweeted
I am very proud to announce that my PhD paper finally came out in Cell! In this *very* collaborative study, we develop and release a deep-learning approach to predict neuron type identity from their electrical signature. doi.org/10.1016/j.cell.2025.… 1/16 🧵
Our “Neuro-AI” paper has just been published in Cell. It describes a new deep learning-based classifier for identifying different cell types in Neuropixels recordings: doi.org/10.1016/j.cell.2025.…
