# State-Reification Networks: Improving Generalization by Modeling the Distribution of Hidden Representations

## TL;DR

State-Reification Networks enhance neural network generalization by modeling and projecting hidden states onto learned distributions, improving robustness especially with limited data and against adversarial attacks.

## Contribution

This paper introduces state reification, a novel method that models hidden state distributions to improve neural network generalization and robustness.

## Key findings

- Better generalization with sparse data
- Improved robustness against adversarial examples
- Effective projection of hidden states onto learned distributions

## Abstract

Machine learning promises methods that generalize well from finite labeled data. However, the brittleness of existing neural net approaches is revealed by notable failures, such as the existence of adversarial examples that are misclassified despite being nearly identical to a training example, or the inability of recurrent sequence-processing nets to stay on track without teacher forcing. We introduce a method, which we refer to as \emph{state reification}, that involves modeling the distribution of hidden states over the training data and then projecting hidden states observed during testing toward this distribution. Our intuition is that if the network can remain in a familiar manifold of hidden space, subsequent layers of the net should be well trained to respond appropriately. We show that this state-reification method helps neural nets to generalize better, especially when labeled data are sparse, and also helps overcome the challenge of achieving robust generalization with adversarial training.