What Can Neural Networks Reason About?

By:

Keyulu Xu (1), Jingling Li (2), Mozhi Zhang (2), Simon S. Du (3), Ken-ichi Kawarabayashi (4), Stefanie Jegelka (1)

MIT
University of Maryland
Institute for Advanced Study (IAS)
National Institute of Informatics (NII)

Conference:

ICLR 2022

The Transcendentals:

The True:

Develops a formal framework to define similarity between task and network structure.

The Good:

The Beautiful:

Other important details: