Artificial intelligence is the most important technological revolution and driving force in the next decade, playing an increasingly important role in all walks of life. It and the development of big data complement each other, not only pushing human society into a smarter world, but also bringing inestimable value to the application of data.





Flink Forward Asia 2019 will focus on Flink’s new technologies and applications in machine learning from November 28 to 30.

  • How far does Flink machine learning go?
  • How do you integrate Flink with frameworks like TensorFlow?
  • What are the practical applications of Flink in machine learning?
To present you Flink machine learning specific application practice and the latest technology landing cases.

Deep Learning On Apache Flink

Jackie chan | Tencent Engineer

This talk will introduce a new Flink project dlonflink. It gives Flink the ability to run distributed training and real-time update model services. Not only can you develop in Scala, but you can run TensorFlow on Flink with just a few lines of Python. It also supports PyToch and MXNet. We’ll detail the design of data exchange between Java and Python and the implementation of framework management.

The architecture and practice of big data analysis and deep learning model inference using Analytics-Zoo on Flink

Senior software architect Shi Dongjie | Intel

Senior Software Architect, Intel. He has been engaged in enterprise computing, risk control, big data analysis, cloud computing container choreography, data analysis and artificial intelligence research and development for many years. He is one of the contributors of Intel open source framework BigDL and Analytics-Zoo. The main contents to be shared are as follows:

  1. Analytics-Zoo: Big data Analytics +AI platform based on Apache Spark, Tensorflow, Keras and BigDL.
  2. Deploy deep learning applications on big data to production environment, support Flink stream processing scenarios and use OpenVINO acceleration.
  3. A Cluster Serving architecture based on message subscription and Flink stream processing.

Apache Flink + TensorFlow, Ctrip real-time intelligent exception detection platform practice

Guo-qing pan | ctrip big data research and development manager

With the development of recent years, real-time computing technology is becoming more and more mature, real-time computing scenarios are becoming more and more diverse, Flink has gradually become one of the first choice of real-time computing engine, from simple real-time ETL to complex CEP scenarios Flink can well control.

This sharing mainly introduces how Ctrip builds ctrip real-time intelligent anomaly detection platform based on Flink and Tensorflow, which is used to solve many problems such as low accuracy, low timeliness, complex rule configuration and labor consumption of rule alarm system, and realize millisecond delay and intelligent detection of business indicators. Based on Flink, a strong fault-tolerant mechanism is implemented.

Machine learning algorithm platform based on Apache Flink practice and open source

Yang xu | alibaba senior expert algorithm

Alibaba Computing Platform Business Division is working with Flink community to open a machine learning algorithm library derived from research, based on which users can more easily build high-performance Flink machine learning jobs. We want to promote the Flink community in the field of machine learning through open source. We also welcome more developers to work with us to build a more powerful and complete Library of Flink algorithms.

This sharing mainly focuses on the technical accumulation and harvest of the team’s development of high-performance machine learning algorithm library based on Flink, as well as the progress of open source.

Apache Flink AI Ecosystem Effort

Chen e ultra | alibaba technical experts Gao Yun | alibaba technical experts

Flink is a distributed computing engine that supports batch data processing. In the use of artificial intelligence in the actual production scene, Flink in engineering including features, online learning, online prediction has some unique advantages, in order to better support the use of artificial intelligence, community, and the various ecological vendors have done some work, this presentation will introduce some recent Flink progress in ecological system of artificial intelligence.

  1. Flink ML Pipline: Defines an API for running machine learning workflows on Flink, including feature engineering, model training, model prediction, etc.
  2. Alink is a machine learning algorithm library based on Flink.
  3. Flink ML Workflow: Based on Flink and combined with the deep learning framework TensorFlow/PyTorch, Flink ML Workflow provides users with an API that connects the various phases of a machine learning Workflow and supports both batch and stream operation modes.
  4. Streaming-based Mini-Batch Iteration Architecture: Flink’s new architecture design supports iteration in Streaming operation mode.






The original link

This article is the original content of the cloud habitat community, shall not be reproduced without permission.