CNT (Conditioning on Noisy Targets): A new Algorithm for Leveraging Top-Down Feedback

TL;DR

CNT introduces a regularizer that conditions models on noisy targets at varying noise levels, enabling better learning by focusing on simpler sub-problems and gradually tackling harder examples, improving supervised learning.

Contribution

The paper presents a novel regularizer called Conditioning on Noisy Targets (CNT) that leverages noisy targets at different noise levels to enhance supervised learning.

Findings

Improves learning by focusing on easier examples first.

Enables models to handle noisy labels more effectively.

Facilitates gradual learning from simple to complex tasks.

Abstract

We propose a novel regularizer for supervised learning called Conditioning on Noisy Targets (CNT). This approach consists in conditioning the model on a noisy version of the target(s) (e.g., actions in imitation learning or labels in classification) at a random noise level (from small to large noise). At inference time, since we do not know the target, we run the network with only noise in place of the noisy target. CNT provides hints through the noisy label (with less noise, we can more easily infer the true target). This give two main benefits: 1) the top-down feedback allows the model to focus on simpler and more digestible sub-problems and 2) rather than learning to solve the task from scratch, the model will first learn to master easy examples (with less noise), while slowly progressing toward harder examples (with more noise).