Learning to Learn from Weak Supervision by Full Supervision

TL;DR

This paper introduces a method for training neural networks with large amounts of weakly labeled data by using a confidence network trained on true labels to control gradient updates, improving learning quality.

Contribution

It presents a novel dual-network approach combining weak supervision with a confidence network to enhance neural network training.

Findings

Effective control of gradient updates using confidence scores.

Improved model performance with weakly labeled data.

Robustness against noisy labels.

Abstract

In this paper, we propose a method for training neural networks when we have a large set of data with weak labels and a small amount of data with true labels. In our proposed model, we train two neural networks: a target network, the learner and a confidence network, the meta-learner. The target network is optimized to perform a given task and is trained using a large set of unlabeled data that are weakly annotated. We propose to control the magnitude of the gradient updates to the target network using the scores provided by the second confidence network, which is trained on a small amount of supervised data. Thus we avoid that the weight updates computed from noisy labels harm the quality of the target network model.