Gabriele Di Antonio

This paper introduces a recurrent neural network model with heterogeneous self-interactions that generates scale-invariant memory and cognitive flexibility without requiring criticality.

This study shows that the premotor cortex encodes a mental line for transitive inference, with neural activity reflecting symbolic distances and learning through geometric transformations.

This paper uses reservoir computing with noisy linear recurrent neural networks to create digital twins of brain activity from resting-state fMRI data, enabling prediction and decomposition of BOLD signals into low-dimensional latent states.

The paper proposes a geometric model for serial learning that explains transitive inference behavior in animals and motor decision-making in recurrent networks.