Interview with Xie Liang, Microsoft Chief Data Scientist: From Economics to AI and Cloud Computing

Xie Liang, chief data scientist at Microsoft Headquarters, earned his undergraduate degree in economics from Southwestern University of Finance and Economics, worked in credit assessment at Industrial and Commercial Bank of China, then pursued applied econometrics at SUNY. His research interests include mixed models, data mining, and leveraging SAS, and he authored "Keras Quick Start: Practical Deep Learning with Python".

With over a decade of machine learning experience, he transitioned from economics to become a top data scientist in cloud computing. In a recent CSDN interview, he explained why he was attracted to machine learning, the challenges he faced, and how he deepened his expertise in deep learning.

He noted that econometrics and machine learning share many statistical foundations, making the shift natural. At Microsoft Azure, he worked on both SaaS customer‑behavior analytics and IaaS infrastructure optimization, using large‑scale data to improve resource allocation, fault prediction, and capacity planning.

According to Xie, any business with abundant data that can reveal patterns for automation and optimization can benefit from machine learning. Azure provides the necessary scale and competitive pressure to adopt intelligent, automated operations, reducing incidents such as node‑failure‑related capacity loss by about 30%.

He believes future DevOps teams must become intelligent to lower developer workload and increase system efficiency, recommending dedicated data‑mining teams and early integration of AI methods into development pipelines.

When choosing deep‑learning frameworks, Xie emphasizes practical considerations: the tool that can quickly move a concept to production is preferred, and while Microsoft has its own platforms, open‑source options like Keras are also embraced.

Regarding his book, Xie explains that many practitioners wanted a concise, hands‑on guide to start with Keras without steep theory, resulting in a recipe‑style manual that focuses on practical implementation rather than deep theory.

He outlines three stages of applying machine learning in industry: front‑end problem translation, middle‑end algorithm implementation, and back‑end result interpretation and business integration. He warns that while tools simplify the middle stage, they raise expectations for the front and back ends.

For readers seeking deeper knowledge, Xie recommends classic deep‑learning textbooks such as Goodfellow and Bengio's "Deep Learning" and staying current with top conference papers.

He also addresses practical concerns: Keras compilation time is negligible compared to training, and while Keras suffices for light production workloads, heavy‑duty scenarios may require direct TensorFlow or CNTK APIs.

Finally, Xie will present a talk titled "Building Your Own Deep‑Learning Model from Scratch with Keras" at the SDCC 2017 AI Technology Summit on October 28, alongside other AI experts from Alibaba, SenseTime, and more.