Machine learning has had substantial impact on many fields, and with this so to does AutoML. By making machine learning more efficient, this knock on effect accelerates the practical applications and utility that machine learning can provide to domain-experts.
In fields with tabular data, AutoML has been crucial in finding suitable pipelines within a reasonable time frame, allowing domain-experts to focus more on what they know, and less time tinkering with various machine learning models. With theoretical frameworks for multi-objective and constrained optimization, AutoML empowers the practitioner to see the often unknown trade-offs in their metrics that exist when applying machine learning to their data. One such metric of increasing importance is fairness. While no one definition suffices, fairness-aware AutoML frees stake-holders from the minutia of optimization and focus on the real-world impact of their models. One recent highlight in the field of AutoML for tabular data is the TabPFN, a state-of-the-art model that requires no training out of the box, enabling a host of new applications, but, that is flexible enough to enable domain-experts to encode their own beliefs about their data.
One other area AutoML is heavily applied to is RL. Under construction…
An area of concern is that machine learning is expensive, and optimizing on top of that is even more so. There is a growing area of sustainable AutoML, that looks at how we can lessen the carbon footprint of such expensive procedures, and even find smaller, more energy efficient models for use on edge devices but also to reduce the energy consumption of machine learning in general.
AutoML has had some great success in practical applications too, notably RNA design, EEG data and throughout Healthcare. Under construction…