A Computability Perspective on (Verified) Machine Learning
This work provides a foundational perspective on verified ML, addressing a conceptual gap for researchers in formal verification and machine learning.
The paper tackles the problem of defining verified machine learning by using computable analysis to establish that the computational tasks underlying verified ML are computable in principle.
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.