Samuel Lippl

This paper reframes transitive inference as a linear classification problem using least-squares estimation, showing it can reproduce key empirical features of TI without explicit assumptions about transitivity.

The paper develops a mathematical framework to explain how learning models, including neural networks, can perform transitive inference through statistical principles and representation learning.

This paper investigates whether iterative convergent computation improves ResNets but finds it does not enhance classification accuracy on standard or challenging visual tasks.