Aura - a Machine Learning Platform for Data-driven Applications

With high hope for the recent tech revolution driven by Big Data and Artificial Intelligence (AI), businesses across verticals have been investing heavily in deploying big data technologies, while expecting significant improvements in every aspect of their operation. However, in many domains, real progress and impacts have been slow and lukewarm despite the availability of infrastructure hardware and software supports. As the largest Big Data technology provider in China, our experiences from deep interactions with many businesses allow us to identify a few important gaps that undermine the progresses in their adoption of predictive analytics and data-driven operations. First of all, building effective data-driven applications would require both domain knowledge and data science skills. However, the human resources with skills sufficient for conducting machine-learning modeling and data science analytics are still relatively scarce inside traditional business regime. On the other hand, technology providers have limited insight of the domain-specific problems and unique characteristics of data sets in different verticals, despite their expertise in general data science methodology. Secondly, the uncertainty of developing data-driven applications is high, as businesses always find themselves stepping into new territories that require lot of explorations and try-and-errors. The cost for obtaining relevant and high-quality data with clear semantics is considerable. And there is no guarantee on the existence of relevant signals for building predictive models inside the business data warehouse. Last but not least, given the fast pace of innovation in technologies like Deep Learning, it is hard for business to pick and choose the most suitable stack of tools for their business needs, let alone taking advantage of the cutting edge. To address these gaps, we designed and developed AURA, a machine learning powered platform for efficient data-driven application development. AURA improve the state-of-the-art by first emphasizing the preservation, standardization and reuse of industrial domain knowledge using its Common Data Model. AURA provides a one-stop system with cutting-edge data science technologies. By integrating the multiple machine learning and deep learning frameworks on top of powerful big data platforms like Spark, AURA enable businesses to choose the most appropriate machine learning algorithms to solve their problems at hand. AURA also provides two ways for user to interact with it: (a) a powerful notebook for data scientist or analyst with advanced programming skills and (b) a visual pipeline that guides intermediate users through common data analytics and modeling scenarios. Varies modeling pipeline templates have been provided, which are designed to guide user with scientific methods and known best practices. Most importantly, on the AURA platform we offer a wide range of tried-and-true data-driven applications for a variety of industries, which can be directly (or with minor tweaks) applied to address business problems out of the package. In this talk, we will also show cases some of the successful applications on AURA in telecom, banking, healthcare, which helps the business to lift revenue, fight fraud, improve customer service, or even developing new products.

Dr. Yu is the CTO of AsiaInfo Data, which provides big data and AI solutions to all three telecom carriers in China. Its distributed big data platform processes over 7PB of data daily. Before joining AsiaInfo, Dr. Yu served as VP Engineering and Chief Architect for Mafengwo, the largest online travel community in China, with over 100 mobile and online million users. Previously, he was VP Engineering and Chief Architect at OpenX, the second largest global ad exchange (next to Google), performing 50 billion RTB transactions per day. He was responsible for the company’s data strategy, mobile product line, and overall architecture, consisting of more than 6000 servers and 15PB of data distributed in 5 global data centers. Dr. Yu is also a serial entrepreneur, co-founded two startups, Portaura in social mobile big data and Martsoft in e-commerce search engine. Early in his career, he spent a number of years in HP(DEC) Systems Research Center, one of the top research labs in the world, where he worked closely with numerous Turing Award recipients in browser technology, search engine, multimedia, and distributed file systems. Dr. Yu holds PhD in Computer Science and Engineering from UNSW and BS in Computer Sciences and BA in Mathematics from UT Austin. He has published papers and gave keynote speeches in numerous major international conferences.