Biggio, Luca
13 publications
ICML
2025
Counting in Small Transformers: The Delicate Interplay Between Attention and Feed-Forward Layers
NeurIPS
2025
On the Bias of Next-Token Predictors Toward Systematically Inefficient Reasoning: A Shortest-Path Case Study
NeurIPSW
2023
Harnessing Synthetic Datasets: The Role of Shape Bias in Deep Neural Network Generalization