Augmenting Neural Networks with First-order Logic

Tao Li, Vivek Srikumar

Today, the dominant paradigm for training neural networks involves minimizing task loss on a large dataset. Using world knowledge to inform a model, and yet retain the ability to p…