Federico Barbero

Attention Sinks and Compression Valleys in LLMs are Two Sides of the Same Coin

Wavelet-Induced Rotary Encodings: RoPE Meets Graphs

Bundle Neural Network for message diffusion on graphs

On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning

Generating Creative Chess Puzzles

Why do LLMs attend to the first token?

Round and Round We Go! What makes Rotary Positional Encodings useful?

softmax is not enough (for sharp out-of-distribution)

Softmax is not Enough (for Sharp Size Generalisation)