Microsoft Investigator Fellow Dr. Guillermo Sapiro at Duke University uses Azure for big data management

Customizing cloud infrastructure

Dr. Sapiro said one of the most significant benefits of using Azure’s cloud computing infrastructure is that it already includes scalable tools that can be used to support everything from clinical analytics to machine learning. Rather than investing precious financial and information technology resources into an on-premises data center, Duke researchers can securely access, manage, and analyze protected patient data in a high-performance cloud computing environment.

“Microsoft, like many companies, has extraordinary teams developing tools,” said Dr. Sapiro. “That doesn't mean that the community doesn't need to develop their own tools, because every researcher has specific needs. But, in the ideal world, I don't need to hire a person or gain expertise using cloud tools. I'm going to use what Azure already provides.”

Those tools include the ability to build machine learning models that can analyze images, comprehend speech, and help make predictions by analyzing massive amounts of data. The ability to aggregate and analyze data from documents, images, and other media helps medical researchers like Dr. Sapiro accelerate scientific discovery and develop innovative solutions, such as new drug treatments, more accurate diagnostics, and customized treatment protocols.

Putting the Azure solution set to work

Dr Sapiro’s team is using Azure Machine Learning to build, train, and deploy machine learning models for their research. With Azure Cognitive Services, they’re able to easily add AI capabilities into the data models and processes. Cloud computing, according to Dr. Sapiro, enables researchers to focus their time and resources on scientific pursuits rather than software development.

For example, developing and maintaining the level of encryption required to protect patient data on-site requires a team of data scientists and IT experts to build and maintain. “Developing tools for encryption may sound simple,” said Dr. Sapiro, “but it's complicated. If I were to do that by myself, it might take an engineer six months, including a lot of money and a lot of servers,” he said. “But Azure already has those encryption tools. Duke University shouldn't be worried about developing our own encryption tools, we should be using the encryption tools that are provided to us.”

One example is how the power of cloud computing is helping the Duke researchers extract data from video. A core problem when it comes to video analysis is determining who is speaking. Automating the process of analyzing the video requires diarization, which parses the data based on speaker identity. Rather than investing in new resources with expertise in diarization, Dr. Sapiro was able to leverage the Azure functionality. “Azure has a diarization algorithm out there," he said. “Boom. We just saved months or years because we don't have to worry about that anymore. My student that was working on an interesting video analysis problem tested it, and it gave us the output we needed.”

The ability to customize an existing diarization algorithm improves Dr. Sapiro’s ability to, for example, develop computational behavioral phenotyping models, which are derived from behavioral data. “Where the machine learning and computer vision challenges are very significant,” he said, “and not smaller than the privacy ones, combining our expertise and data with Microsoft’s capabilities in Azure, and beyond, is a perfect match.” 

Supporting academic research

Microsoft is helping academic researchers like Dr. Sapiro benefit from computational resources in new and innovative ways. Dr. Sapiro is one of 15 research fellows in the Microsoft Investigator Fellowship program, a two-year program that began in fall 2019 and is one of several Microsoft initiatives designed to accelerate research. Other Microsoft programs include AI for Health—which provides grants that support using AI to address global health challenges—and AI for Earth, which offers funding for research into how AI can change the way people and organizations monitor, model, and manage the Earth’s natural systems. In addition, the Microsoft Research Academic Programs support research opportunities for faculty, students, and collaborating institutions.

“What is really needed in universities is support from companies like Microsoft that allows faculty to do open-ended fundamental research for which the near-term impact is unclear,” said Dr. Lawrence Carin, currently Provost at King Abdullah University of Science and Technology (KAUST) in Thuwal, Saudi Arabia. Dr. Carin helped Microsoft launch the Investigator Fellowship program while serving as Vice President for Research at Duke University in Durham, North Carolina. 

“What happens when you do that type of research,” added Dr. Carin, “is sometimes there are surprises, unexpected things that you could never come up with from short-term, myopic goals.”