Democratizing Machine Learning: Contributions in AutoML and Fairness
MCML Authors
Florian Pfisterer
Dr.
* Former Member
Abstract
Florian Pfisterer
Dr.
* Former Member
Abstract
This thesis focuses on democratizing access to machine learning (ML) by improving automated machine learning (AutoML) systems and making ML tools more accessible to non-experts. Key contributions include methods to accelerate hyperparameter optimization by learning from previous experiments, the integration of fairness considerations in AutoML, and the development of software packages such as mlr3pipelines for creating machine learning pipelines and mlr3fairness for auditing and debiasing models. The thesis also includes tools for estimating and mitigating model fairness, such as the mcboost package for multi-calibration, addressing both the technical and ethical challenges of widespread ML deployment. (Shortened.)
BibTeXKey: Pfi22