How Can Businesses Use AutoML
Editor’s Note: Automated Machine Learning is rapidly making the benefits of AI available to companies without fleets of data scientists and Machine Learning Engineers. In this post Wee Hyong Tok, Principle Data Science Manager at Microsoft, explains how Automated ML works and how it can empower businesses to do more.
By using machine learning — fueled by data — businesses can engage their customers, optimize operations, empower employees, and fundamentally change and transform products. Automated machine learning will be a catalyst to empower businesses to enable faster business value realizations.
Automated machine learning enables teams to spontaneously create machine learning pipelines and train machine learning models. This helps both nonexperts and experts in the organization work toward the goal of building innovative machine learning solutions. In this report, you will learn how automated machine learning can jump-start adoption of machine learning in your organization.
What Is Automated Machine Learning?
As described by AutoML.org, automated machine learning “provides methods and processes to make machine learning available for non–machine learning experts, to improve efficiency of machine learning, and to accelerate research on machine learning.” In short, automated machine learning empowers non–machine learning experts to get started with machine learning quickly by automating many of the tasks that data scientists perform each day.
Figure 1 shows some of the common tasks handled by automated machine learning. These tasks include the selection of machine learning algorithms, feature engineering, and selecting the best way to tune the machine learning algorithm (commonly known as hyperparameter tuning).
This is represented in a circle, as shown in the diagram, because it is an iterative process, where automated machine learning iterates through different combinations of algorithms, hyperparameter values, and many more tasks to deliver a ranked list of machine learning pipelines, and evaluates them based on the desired performance metric (e.g., accuracy, precision, AUC, or F1 score).
Common tasks handled by automated machine learning
As businesses explore the use of automated machine learning, it is natural to consider the risks and liabilities of using automatically selected machine learning pipelines and models. For example, will the machine learning pipelines and models be black boxes, which are hard to maintain and interpret? In some regulated industries, it is also important to explain predicted results.
The machine learning pipelines and models that are identified by automated machine learning should be evaluated and reviewed by the same company processes that govern the deployment of AI solutions in the company, ensuring compliance with company policies as well as regulations for the specific industry. To build trust, many of the existing techniques for model interpretability can be used in both the training and inference phases.
As data scientists explore the use of automated machine learning, a common question that one might ask is: are there any reasons data scientists should not use automated machine learning? The answer is: it depends.
Data scientists who prefer maximum flexibility in performing feature engineering (and who prefer choosing and tuning the machine learning algorithms) might prefer to develop the code without the aid of automated machine learning. In some cases, it may be necessary to use specialized machine learning algorithms not provided by automated machine learning.
Three Reasons Why Automated Machine Learning Matters to Businesses
Automated machine learning empowers businesses to achieve more by providing a head start for anyone who wants to get started with building AI solutions. As Google’s Jeff Dean pointed out in his talk at the 2019 International Conference on Machine Learning (ICML), “Millions of organizations worldwide have machine problems (most don’t even realize this yet) but only tens or hundreds of thousands of people trained to solve these problems” (https://bit.ly/3dQ4pzX).
Reducing the time to market for AI solutions matters
By reducing the time taken to develop and train machine learning models, AI solutions can be prototyped faster, demonstrate proof of value quickly, and be hardened and improved as they make their way to production deployment.
Empowering data scientists to achieve more matters
Automated machine learning not only empowers nonexperts but also benefits experienced data scientists. Data science teams will be more productive and efficient, as automated machine learning helps data scientists quickly work through various machine learning algorithms and tuning solutions. This enables data scientists to quickly achieve a good baseline and get a head start toward the best models.
Making machine learning simpler for everyone in the organization matters
Automated machine learning helps make all machine learning simpler for everyone in the business (from data scientists to analysts to anyone in the organization interested in using machine learning to solve business problems). It intelligently figures out the best ways to prepare the data for machine learning, select the machine learning algorithms, tune the hyperparameter values for the algorithms, and optimize the end-to-end machine learning pipeline for the performance metrics that you want to use for model evaluation.
Learn faster. Dig deeper. See farther.
Join the O’Reilly online learning platform. Get a free trial today and find answers on the fly, or master something new and useful.
Wee Hyong is a Principal Data Science Manager with the AI Platform team at Microsoft. He leads the Engineering and Data Science team for the AI for Earth program. Wee Hyong has worn many hats in his career — developer, program/product manager, data scientist, researcher, and strategist, and his track record of leading successful engineering and data science teams has given him unique super powers to be a trusted AI advisor to many customers.