As part of Azure Machine Learning service general availability, Microsoft has a new automated machine learning (automated ML) capabilities. Automated ML allows you to automate model selection and hyperparameter tuning, reducing the time it takes to build machine learning models from weeks or months to days, freeing up more time for them to focus on business problems. The making of automated ML was driven by our commitment to improve the productivity of data scientists and democratize AI. By simplifying machine learning, automated ML enables domain experts in the businesses to rapidly build and deploy machine learning solutions.
Automating the construction and tuning of machine learning (ML) models has long been one of the goals of the ML community. This is due to several factors, most notably a sharp increase in the demand for tailored AI solutions, a relative scarcity of trained ML scientists, and the development of deep learning models with complex architectures requiring accurate design and fine-tuning. Existing automated machine learning (AutoML) techniques have been remarkably successful in identifying good parameters for a given model, sometimes even outperforming humans. However, these options either take too long to train or they work for only a handful of parameters. Our research centres on creating AutoML techniques that outperform existing approaches.
Microsoft solution to this problem leverages thousands of experiments performed in hundreds of different datasets. Now running in preview in Azure machine Learning services, Microsoft approach runs a few models with hyperparameters tuned various ways on a user’s new dataset, to learn the accuracy of the pipeline’s predictions. That information informs the next set of recommendations, over hundreds of iterations.
When compared with other automated approaches, Microsoft method outperforms them in terms of classification accuracy by 2 to 200 times, depending on the task. Even against human data scientists in Kaggle competitions, Microsoft approach often beats 95 percent of the data scientists competing.
