Label-Free Supervision of Neural Networks with Physics and Domain Knowledge

TL;DR

This paper presents a novel method for supervising neural networks using domain knowledge and physical constraints, reducing the need for labeled data in computer vision tasks.

Contribution

It introduces a constraint-based supervision framework that leverages prior knowledge, such as physical laws, instead of relying on labeled examples.

Findings

Effective training of neural networks without labeled data

Significant reduction in labeled data requirements

Applicable to real-world and simulated tasks

Abstract

In many machine learning applications, labeled data is scarce and obtaining more labels is expensive. We introduce a new approach to supervising neural networks by specifying constraints that should hold over the output space, rather than direct examples of input-output pairs. These constraints are derived from prior domain knowledge, e.g., from known laws of physics. We demonstrate the effectiveness of this approach on real world and simulated computer vision tasks. We are able to train a convolutional neural network to detect and track objects without any labeled examples. Our approach can significantly reduce the need for labeled training data, but introduces new challenges for encoding prior knowledge into appropriate loss functions.