PRESC Classifier Copies Package

The ubiquitous use over the Internet, and in particular in search engines, of often proprietary black-box machine learning models and APIs in the form of Machine Learning as a Service, makes it very difficult to control and mitigate their potential harmful effects (such as lack of transparency, privacy safeguards, robustness, reusability or fairness). Machine Learning Classifier Copying allows us to build a new model that replicates the decision behaviour of an existing one without the need of knowing its architecture nor having access to the original training data. A suitable copy allows to audit the already deployed model, mitigate its shortcomings, and even introduce improvements, without the need to build a new model from scratch, which requires access to the original data.

This project aims to implement a practical solution of this innovative technique into PRESC, an existing free software tool for the evaluation of machine learning classifiers, so that classifier copies are automated and can be easily created by developers using machine learning, in order to reuse, evaluate, mitigate and improve black-box models, ensure a personal data privacy safeguard into their machine learning models, or for any other application.