Supervised Encoding for Discrete Representation Learning

TL;DR

This paper introduces SEQ, a supervised encoding model that uses a quantizer to create interpretable clusters representing data styles, enabling style transfer and better feature understanding.

Contribution

The paper proposes a novel supervised encoding framework with a quantizer that enhances interpretability and supports style transfer between classes.

Findings

Quantizer creates interpretable class clusters.

Decoder enables style transfer between subclasses.

Model improves understanding of feature representations.

Abstract

Classical supervised classification tasks search for a nonlinear mapping that maps each encoded feature directly to a probability mass over the labels. Such a learning framework typically lacks the intuition that encoded features from the same class tend to be similar and thus has little interpretability for the learned features. In this paper, we propose a novel supervised learning model named Supervised-Encoding Quantizer (SEQ). The SEQ applies a quantizer to cluster and classify the encoded features. We found that the quantizer provides an interpretable graph where each cluster in the graph represents a class of data samples that have a particular style. We also trained a decoder that can decode convex combinations of the encoded features from similar and different clusters and provide guidance on style transfer between sub-classes.