MLOps—the path to building a competitive edge

Enterprises today are transforming their businesses using Machine Learning (ML) to develop a lasting competitive advantage. From healthcare to transportation, supply chain to risk management, machine learning is becoming pervasive across industries, disrupting markets and reshaping business models.

Organizations need the technology and tools required to build and deploy successful Machine Learning models and operate in an agile way. MLOps is the key to making machine learning projects successful at scale. What is MLOps ? It is the practice of collaboration between data science and IT teams designed to accelerate the entire machine lifecycle across model development, deployment, monitoring, and more. Microsoft Azure Machine Learning enables companies to fully embrace MLOps practices will and truly be able to realize the potential of AI in their business.

One great example of a customer transforming their business with Machine Learning and MLOps is TransLink. They support Metro Vancouver's transportation network, serving 400 million total boarding’s from residents and visitors as of 2018. With an extensive bus system spanning 1,800 sq. kilometers, TransLink customers depend heavily on accurate bus departure times to plan their journeys.

To enhance customer experience, TransLink deployed 18,000 different sets of Machine Learning models to better predict bus departure times that incorporate factors like traffic, bad weather, and other schedule disruptions. Using MLOps with Azure Machine Learning they were able to manage and deliver the models at scale.

“With MLOps in Azure Machine Learning, TransLink has moved all models to production and improved predictions by 74 percent, so customers can better plan their journey on TransLink's network. This has resulted in a 50 percent reduction on average in customer wait times at stops.”–Sze-Wan Ng, Director of Analytics & Development, TransLink.

Johnson Controls is another customer using Machine Learning Operations at scale. For over 130 years, they have produced fire, HVAC and security equipment for buildings. Johnson Controls is now in the middle of a smart city revolution, with Machine Learning being a central aspect of their equipment maintenance approach.

Johnson Controls runs thousands of chillers with 70 different types of sensors each, streaming terabytes of data. MLOps helped put models into production in a timely fashion, with a repeatable process, to deliver real-time insights on maintenance routines. As a result, chiller shutdowns could be predicted days in advance and mitigated effectively, delivering cost savings and increasing customer satisfaction.

“Using the MLOps capabilities in Azure Machine Learning, we were able to decrease both mean time to repair and unplanned downtime by over 66 percent, resulting in substantial business gains.”–Vijaya Sekhar Chennupati, Applied Data Scientist at Johnson Controls

Getting started with MLOps

To take full advantage of MLOps, organizations need to apply the same rigor and processes of other software development projects.

To help organizations with their machine learning journey, GigaOm developed the MLOps vision report that includes best practices for effective implementation and a maturity model.

Maturity is measured through five levels of development across key categories such as strategy, architecture, modeling, processes, and governance. Using the maturity model, enterprises can understand where they are and determine what steps to take to ‘level up’ and achieve business objectives.

 

Building MLOps maturity

 

“Organizations can address the challenges of developing AI solutions by applying MLOps and implementing best practices. The report and MLOps maturity model from GigaOm can be a very valuable tool in this journey,”– Vijaya Sekhar Chennupati, Applied Data Scientist at Johnson Controls.

To learn more, read the GigaOm report and make machine learning transformation a reality for your business.

More information

Source: Azure Blog Feed

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.