Intel Presents Analytics Zoo at AI Field Day 1

Intel’s Transformation for the Intelligence Era

Watch on Vimeo
Watch on YouTube

Fadi Zuhayri, Sr. Director, Big Data AI, introduces Intel’s transformation for the the intelligence era. He begins with some background, as compute democratization has brought technology to the exascale era. He also presents the drivers for transformation, 5G, artificial intelligence (AI), and the intelligent edge, and how these have driven the technology pillars addressed by Intel. Intel’s analytics and AI strategy focuses on diverse hardware, software, and ecosystems from edge to cloud, providing a foundation for AI workloads. Next, he looks at Intel’s oneAPI which provides an abstraction layer to support any application or middleware on any hardware, and Intel DevCloud, which makes it easier to learn and develop software.

Personnel: Fadi Zuhayri

Intel Analytics Zoo: Software Platform for Big Data AI

Watch on Vimeo
Watch on YouTube

Jason Dai, Senior Principal Engineer, presents Intel’s Analytics Zoo, a software platform for big data AI. Just as BigDL enables distributed, high-performance deep learning, Analytics Zoo is a unified big data AI platform for TensorFlow, PyTorch, Keras, BigDL, OpenVINO, Ray, and Apache Spark. This open source project reflects the transformation of big data, which now includes AI. After a brief case study, Dai presents the overall architecture of Analytics Zoo and discusses seamless scaling of analytics and AI from laptop to distributed big data.

Personnel: Jason Dai

Intel Analytics Zoo Technical Overview and Case Studies

Watch on Vimeo
Watch on YouTube

Jason Dai, Senior Principal Engineer, gives a deeper technical overview of Intel’s Analytics Zoo, an open source software platform for big data AI. Analytics Zoo uses distributed TensorFlow/PyTorch on Apache Spark, and Dai discusses how it is used in network quality prediction by SK Telecom. Next he focuses on RayOnSpark to run Ray programs directly on the big data platform, giving the example of Burger King’s fast food recommendation engine which leverages an end-to-end training pipeline with RayOnSpark. Turning to the ML workflow question, Dai discusses how a scalable AutoML allows time series prediction, using the example of Tencent Cloud’s TI-One ML platform. Finally, Dai presents the Zouwu open source time series framework on Analytics Zoo.

Personnel: Jason Dai

Building Big Data AI Applications on Intel Analytics Zoo

Watch on Vimeo
Watch on YouTube

Jason Dai, Senior Principal Engineer, discusses practical applications of Intel’s Analytics Zoo, an open source software platform for big data AI. This presentation focuses on four applications: Recommendation, time series analysis, computer vision, and natural language processing. His first example is the food recommendation engine used by Burger King, which uses a transformer cross transformer (TxT) model leveraging HDFS, Apache Spark and MXNet, and Ray. Next, Dai discusses how SK Telecom’s time series based network quality prediction solution is able to run up to 6x faster using Analytics Zoo on Intel Xeon and presents a similar use case of wind power prediction using Analytics Zoo by GoldWind, which improved accuracy to 79% with a 4x training speedup. When it comes to computer vision, Dai presents industrial inspection by Midea and KUKA and AI-assisted radiology with Dell EMC. Finally, natural language processing (NLP) is presented, using a chatbot in Microsoft Azure and job recommendation engine in Talroo as examples.

Personnel: Jason Dai