When working with machine learning algorithms, there is a lot of tweaking of the models before they are perfect. Automated machine learning, AutoML in short, automates the process of selecting the algorithms and their hyperparameters (algorithm settings) for each specific problem. Usually, machine learning experts with extensive domain knowledge are needed to find the real good solutions. Also, in real-world contexts, selecting the best algorithm and its parameters could be very time-consuming. It may not even be possible based on lack of knowledge of the system. AutoML can help to make machine learning easier for non-machine learning experts.
Ekkono’s AutoML framework selects hyperparameters that are individually tailored for each machine learning algorithm. This process is done by looking at the actual data for each use case. It optimizes the full pipeline architecture and build a solution that is unique for the specific problem. Through a series of optimization techniques, AutoML selects hyperparameters and algorithms that are suited to address the specific problem.
See all research areas.