Interview: Joseph Sirosh, Microsoft on How Azure ML is Simplifying Predictive Analytics
Microsoft Corporate VP of Machine Learning, Joseph Sirosh talks about the recently released Azure ML, users’ feedback, favorite business use cases, how Azure ML fits in the Microsoft’s portfolio for Big Data solutions and more.
Joseph Sirosh recently joined Microsoft from Amazon where he was the VP for the Global Inventory Platform and CTO of the core retail business. In this role he had responsibility for the science and software behind Amazon’s supply chain and order fulfillment systems, as well as the central Machine Learning group which he built and led.
During his 9 years at Amazon, he managed a variety of teams including forecasting, inventory, supply chain and fulfillment, fraud prevention systems, data warehouse and a novel data-driven seller lending business. Prior to Amazon, Joseph worked for Fair Isaac Corporation as VP of R&D. Joseph is passionate about Machine Learning and its applications and has been active in the field since 1990.
Here is my interview with him:
Anmol Rajpurohit: Q1. Congratulations on the successful roll-out of Microsoft Azure Machine Learning. What is the most prominent feedback that you are getting from clients? What features are generating a lot of interest and which ones still need some more work?
Joseph Sirosh: Thank you. Our customers are most excited by how fast they can deploy a machine learning solution. That’s where Azure Machine Learning is really
differentiated. The integrated experience between data sources, modeling, and operationalizing to a web service allows deployments to take hours or days, when they would have previously taken months. For instance, we launched a collaborative effort with Dell Statistica and Azure Machine Learning recently. The Dell team was able to create a working integration with Statistica the same week they met with us and learned about the service for the first time. Customers have also been interested in being able to put R code into production as REST APIs.
The service is production ready today, as evidenced by customers and partners such as Pier 1 Imports, Sumo Logic and Carnegie Mellon. But we are continuously adding new features. For instance, we introduced a machine learning category in the Azure Marketplace recently. Solutions on this marketplace, such as recommendations and forecasting APIs, are built by Microsoft data scientists and our partners and customers. This category in the Marketplace is designed for anyone to be able to post and monetize their solution.
We’re also always working to reduce the barriers to access the service, so we recently made available a free tier that allows users to try Azure Machine Learning with just a Microsoft account ID.
We’re continuously updating Azure Machine Learning, so keep an eye on the machine learning blog for the latest updates and other posts about machine learning at Microsoft.
AR: Q2. Predictive Analytics has a plethora of use cases across industry. Out of all the Azure ML business use cases you have seen so far, which ones are your favorite? Why?
JS: One of my favorite use cases is ThyssenKrupp as it shows the power of machine
learning and how it can impact everyone’s daily lives in positive ways. I take an elevator every day to my office and I don’t think about it -- I just assume it will work flawlessly. For ThyssenKrupp, ensuring millions of elevators worldwide are maintained in top notch condition is a huge opportunity. Through the power of Internet of Things and machine learning, they can predict and prevent an elevator maintenance issues before it even happens so they can service the elevator proactively. This allows them to both reduce maintenance cost and improve customer service -- a combination that is hard to achieve.
AR: Q3. How does Azure ML benefit from Microsoft's extensive portfolio of software products and services?
JS: It’s really our Azure customers that are benefiting from this new capability. With Azure Machine Learning, customers who have SQL Server data in a VM, Azure SQL Database, Azure Table
storage, Azure Blob storage or Azure HDInsight can now easily bring in that data into our modeling studio and perform predictive analytics. This is especially valuable for our Azure HDInsight customers who are managing and seeking to get business value from huge quantities of data. By using Azure Machine Learning to build predictive analytics APIs on the cloud, any application on any device can use predictive intelligence. Users of business intelligence tools such as Power BI and Excel can easily call these APIs -- whether on-premises or on the cloud -- and integrate predictive analytics into dashboards and visualizations.
AR: Q4. Generic Analytics solutions do not deliver a lot of business value, mainly due to the absence of business context. What solutions does the Microsoft Data Platform currently offer for industry-specific Analytics needs? Does the Azure ML offering include any customized solutions?
JS: Azure Machine Learning is focused on empowering data scientists and developers to deliver great business value by operationalizing predictive analytics very easily using their domain knowledge and business context. To enable a rich collection of capabilities for them, we have built open source tools such as R into our solution, with over 400 R packages provided for use in
our product, plus the ability to easily drop in custom R code. The resulting model can then be easily operationalized as a web API for any end application to use, anywhere, for driving business outcomes. We created a video recently that describes how this solution empowers data scientists.
Our partners and customers have taken this ball and run with it, developing custom solutions for predictive maintenance, customer churn, fraud detection, online marketing optimization, and other business challenges. Over time, as we engage with solutions for several industries, we expect to create toolkits that make industry-specific solution building much easier.
Second part of the interview will be published soon.
Related:
