On the Sample Complexity of End-to-end Training vs. Semantic Abstraction Training

TL;DR

This paper compares end-to-end training and semantic abstraction approaches, showing that semantic modularity can drastically reduce the number of training samples needed for high-accuracy tasks like autonomous driving.

Contribution

It provides a theoretical and empirical comparison of sample complexity between end-to-end and modular semantic approaches in high-accuracy scenarios.

Findings

End-to-end training can require exponentially more samples than semantic abstraction.

Semantic modularity reduces sample complexity significantly.

High-accuracy autonomous driving tasks benefit from semantic decomposition.

Abstract

We compare the end-to-end training approach to a modular approach in which a system is decomposed into semantically meaningful components. We focus on the sample complexity aspect, in the regime where an extremely high accuracy is necessary, as is the case in autonomous driving applications. We demonstrate cases in which the number of training examples required by the end-to-end approach is exponentially larger than the number of examples required by the semantic abstraction approach.