Improving Compositionality of Neural Networks by Decoding Representations to Inputs

TL;DR

This paper introduces Decodable Neural Networks (DecNN), which constrain neural activations to decode back to inputs, enhancing compositionality, uncertainty estimation, and robustness while maintaining accuracy.

Contribution

The paper proposes DecNN, a novel neural network design that improves compositionality and uncertainty estimation by jointly training with a decoding mechanism.

Findings

DecNN matches standard neural networks in accuracy.

DecNN improves out-of-distribution and adversarial example detection.

DecNN can be combined with pretrained models for regularization.

Abstract

In traditional software programs, it is easy to trace program logic from variables back to input, apply assertion statements to block erroneous behavior, and compose programs together. Although deep learning programs have demonstrated strong performance on novel applications, they sacrifice many of the functionalities of traditional software programs. With this as motivation, we take a modest first step towards improving deep learning programs by jointly training a generative model to constrain neural network activations to "decode" back to inputs. We call this design a Decodable Neural Network, or DecNN. Doing so enables a form of compositionality in neural networks, where one can recursively compose DecNN with itself to create an ensemble-like model with uncertainty. In our experiments, we demonstrate applications of this uncertainty to out-of-distribution detection, adversarial example detection, and calibration -- while matching standard neural networks in accuracy. We further explore this compositionality by combining DecNN with pretrained models, where we show promising results that neural networks can be regularized from using protected features.