GigaOm Radar for AutoML

Automated machine learning (AutoML) is regarded as a “quiet revolution” in AI. But what makes AutoML revolutionary is not just automation and acceleration of the machine learning process, but also its support for increased accuracy of machine learning models and its democratization of data science. AutoML is definitely in a hype cycle, but it’s also an effective technology that we believe will be integral to a new era in AI.

AutoML has two main goals. The first is to democratize data science so that internally complex models can be utilized by everyone to improve decision-making. To the extent AutoML succeeds here, it can solve the problem of not having enough data scientists to make sense of big data and give everyone a chance to be part of every step of decision-making. The second goal is to automate many repetitive, tedious tasks involved in machine learning. For example, with a manual process, it can take a very long time to try a range of machine learning algorithms in order to find the best one. AutoML automates such tasks, and trains models in parallel, thus maximizing the optimization and increasing the overall accuracy of models an organization may build.

AutoML’s value comes from automating multiple machine learning steps, such as exploratory data analysis, data preparation, feature engineering, model selection, model training, hyperparameter tuning, and deployment. Automating all these steps shortens the traditional data science workflow significantly because users need only upload their data, set a target variable (such as the column to be predicted) and, in most cases, let the AutoML engine do the rest. As such, AutoML not only enables more people to take advantage of machine learning, it also helps experienced data scientists focus on more important strategic steps, like conceptualizing models and carefully reviewing their performance in order to avoid bias. Last but not least, AutoML helps to prevent the human errors that can occur when model optimization is carried out manually.