What Can Neural Networks Reason About?
arXiv: Learning, 2019.
We introduce an algorithmic alignment framework to formalize the interaction between the structure of a neural network and a reasoning process, and provide preliminary results on sample complexity
Neural networks have succeeded in many reasoning tasks. Empirically, these tasks require specialized network structures, e.g., Graph Neural Networks (GNNs) perform well on many such tasks, but less structured networks fail. Theoretically, there is limited understanding of why and when a network structure generalizes better than others, al...More
PPT (Upload PPT)