A Unified Plug-and-Play Framework for Effective Data Denoising and Robust Abstention

TL;DR

This paper introduces a unified framework that improves data quality and prediction reliability in deep neural networks by denoising training data and abstaining from uncertain test predictions without altering existing models.

Contribution

It proposes a data density-based filtering framework that enhances DNN performance and robustness without modifying model architectures or loss functions.

Findings

Outperforms state-of-the-art data denoising methods.

Effectively abstains from uncertain test data.

Works across multiple datasets and CNN architectures.

Abstract

The success of Deep Neural Networks (DNNs) highly depends on data quality. Moreover, predictive uncertainty makes high performing DNNs risky for real-world deployment. In this paper, we aim to address these two issues by proposing a unified filtering framework leveraging underlying data density, that can effectively denoise training data as well as avoid predicting uncertain test data points. Our proposed framework leverages underlying data distribution to differentiate between noise and clean data samples without requiring any modification to existing DNN architectures or loss functions. Extensive experiments on multiple image classification datasets and multiple CNN architectures demonstrate that our simple yet effective framework can outperform the state-of-the-art techniques in denoising training data and abstaining uncertain test data.