A Computability Perspective on (Verified) Machine Learning

TL;DR

This paper explores verified machine learning through computable analysis, establishing a model-agnostic framework that demonstrates the fundamental computability of the underlying tasks, bridging ML verification and theoretical computability.

Contribution

It introduces a computability-based perspective on verified ML, providing a formal, model-agnostic foundation for understanding the computational nature of verification tasks.

Findings

Verified ML tasks are in principle computable

Computable analysis offers a unifying framework for ML verification

Model-agnostic approach broadens applicability of verification methods

Abstract

There is a strong consensus that combining the versatility of machine learning with the assurances given by formal verification is highly desirable. It is much less clear what verified machine learning should mean exactly. We consider this question from the (unexpected?) perspective of computable analysis. This allows us to define the computational tasks underlying verified ML in a model-agnostic way, and show that they are in principle computable.