Building AI to play the FIFA video game using distributed TensorFlow

Shengsheng Huang (Intel), Jason (Jinquan) Dai (Intel)

11:00am–11:40am Thursday, October 31, 2019

Location: Grand Ballroom A/B

Applications

Who is this presentation for?

Senior software architects

Level

Intermediate

Description

Using AI to play games is often perceived as an early step toward achieving general machine intelligence, as the ability to reason and make decisions based on sensed information is an essential part of general intelligence. People have been interested in using AI to play games for quite a while. Instead of working on popular simple simulated game environments like gym Atari, Shengsheng Huang and Jason Dai chose to build a new platform that allows AI agents to play the FIFA video game autonomously. The complexity in the visual and state information and decision making in FIFA requires cooperation of various kinds of technology in computer vision (e.g., object detection, object tracking, optical character recognition (OCR)), reinforcement learning, imitation learning, and meta-learning. And such a real-time 3D video game requires the agent’s speed in processing visual signals and taking actions. Therefore, it’s a very good playground to explore the potential of (distributed) TensorFlow in a complicated setting.

Shengsheng and Jason detail the experiments and insights that come from building various kinds of AI agents to play FIFA using (distributed) TensorFlow on Spark and Analytics Zoo. You’ll see the results, demos, and best practices, and maybe you’ll be inspired to build your own AI agents for such games.

Prerequisite knowledge

A basic understanding of TensorFlow, deep learning, neural networks, reinforcement learning, and object detection

What you'll learn

Discover how distributed TensorFlow can be used to play FIFA

Shengsheng Huang

Intel

Shengsheng (Shane) Huang is a software architect at Intel and an Apache Spark committer and PMC member, leading the development of large-scale analytical applications and infrastructure on Spark in Intel. Her area of focus is big data and distributed machine learning, especially deep (convolutional) neural networks. Previously at the National University of Singapore (NUS), her research interests are large-scale vision data analysis and statistical machine learning.

Shengsheng（Shane）Huang是英特尔的软件架构师，也是Apache Spark的贡献者和PMC成员。她领导着英特尔基于Spark的大规模分析应用和基础架构的开发。她关注的领域是大数据和分布式机器学习，尤其是深度（卷积）神经网络。她之前就读于新加坡国立大学（NUS），研究兴趣是大规模视觉数据分析和统计机器学习。

Website

Jason (Jinquan) Dai

Intel

Jason (Jinquan) Dai is a senior principal engineer and CTO of big data technologies at Intel, where he is responsible for leading the global engineering teams (located in both Silicon Valley and Shanghai) on the development of advanced big data analytics (including distributed machine and deep learning), as well as collaborations with leading research labs (e.g., UC Berkeley AMPLab and RISELab). Jason is an internationally recognized expert on big data, cloud, and distributed machine learning; he is the program cochair of the O’Reilly AI Conference in Beijing, a founding committer and PMC member of Apache Spark, and the creator of BigDL, a distributed deep learning framework on Apache Spark.