Deep Transform: Error Correction via Probabilistic Re-Synthesis

TL;DR

This paper introduces a deep transform method that uses neural networks to correct errors in data by re-synthesizing inputs, effectively rejecting noise and recovering heavily degraded data in an unsupervised manner.

Contribution

The paper presents a novel deep transform approach that enables error correction through probabilistic re-synthesis, without requiring supervised training for specific error types.

Findings

Successfully recovers heavily degraded data

Demonstrates error rejection outside feature space

Effective unsupervised error correction

Abstract

Errors in data are usually unwelcome and so some means to correct them is useful. However, it is difficult to define, detect or correct errors in an unsupervised way. Here, we train a deep neural network to re-synthesize its inputs at its output layer for a given class of data. We then exploit the fact that this abstract transformation, which we call a deep transform (DT), inherently rejects information (errors) existing outside of the abstract feature space. Using the DT to perform probabilistic re-synthesis, we demonstrate the recovery of data that has been subject to extreme degradation.