Fabrizio Silvestri

Composable Sparse Subnetworks via Maximum-Entropy Principle

Directional Sheaf Hypergraph Networks: Unifying Learning on Directed and Undirected Hypergraphs

Energy Guided Smoothness to Improve Robustness

MASS: MoErging through Adaptive Subspace Selection

Evading Overlapping Community Detection via Proxy Node Injection

Double Descent Revisited: When Noise Amplifies and Optimizers Decide

One Size Does not Fit All: Cross-Architectural Layer-wise Representations in Diffusion Models

What are you sinking? A geometric approach on attention sink

ATM: Improving Model Merging by Alternating Tuning and Merging