Twitxy Saved

Alessandro Favero

@alesfav

Physics-AI fellow @Cambridge_Uni explaining the scientific principles behind AI. Formerly @EPFL, @Amazon AI.
Cambridge & London, UK alesfav.github.io Joined September 2020
978 Following 748 Followers 188 Posts

Latest posts 5

Alessandro Favero @alesfav
@MatthieuWyart on @MLStreetTalk on our work:
- How generative models become creative, from polynomially few examples to generating exponentially many novel ones (against Chomsky's poverty of the stimulus)
- Why predicting latents/abstractions 脿 la JEPA beats predicting tokens
Machine Learning Street Talk Verified account @MLStreetTalk Jun 20
> Natural data is "generated" from a constrained hierarchical / compositional function.

> Deep networks learns that hidden structure from polynomially few examples and creatively generate exponentially many valid new ones.

> The depth of the network is what's important to overcome the curse of dimensionality, and potentially invalidate Chomsky's poverty of stimulus argument.

> Prof @MatthieuWyart, a physicist (Johns Hopkins / EPFL) was the senior author of the Random Hierarchy Model.
View quoted post on X
Alessandro Favero @alesfav
Super cool work predicting the scaling exponents of pre-trained LLMs from language statistics alone.
Francesco Cagnetta Verified account @Fraccagnetta Feb 10
馃毃 We derive data-limited neural scaling exponents directly from measurable corpus statistics.

No synthetic data models, only two ingredients:
-decay of token-token correlations with separation;
-decay of next-token conditional entropy with context length.
View quoted post on X
Alessandro Favero @alesfav
馃帗My PhD thesis is now on arXiv! It follows a thread of compositionality in AI: from locality in CNNs, to the 'grammar' diffusion models learn to be creative, to task composition in foundation models.

Officially a Dr 馃帀 and starting as a Physics-AI Fellow at DAMTP @Cambridge_Uni
Stat.ML Papers @StatMLPapers Oct 8, 2025
The Physics of Data and Tasks: Theories of Locality and Compositionality in Deep Learning ift.tt/9B0HFnC
View quoted post on X