GDC + [email protected] 2018 Session Scheduler

View, browse and sort the ever-growing list of GDC sessions by time, pass type, track, and format. With GDC Session Scheduler, you can build your schedule in advance and access it during the show via export or with the GDC Mobile App. Sessions do fill up, so please arrive early to sessions that you would like to attend. Adding a session to your schedule does not guarantee you a seat.

In order to best utilize Session Scheduler, please use the login credentials you've set up on this page as your login credentials on our GDC Mobile App. Your login credentials are the same as your registration login. You can download the app on iTunes or Google Play.

'Race for the Galaxy': A Neural Network in Production

Summit Speaker:

Theresa Duringer (CEO, Temple Gates Games)

Location: Room 2002, West Hall

Date: Tuesday, March 20

Time: 4:40pm - 5:10pm

Pass Type: All Access, GDC Conference + Summits, GDC Summits - Get your pass now!

Topic: AI Summit

Format: Session

Vault Recording: Video

Audience Level: Intermediate

Neural networks have long been viewed with skepticism in game AI. However, there are times when their use is not only appropriate, but a powerful and time-saving approach, particularly for small teams with limited resources. This lecture will explain how 'Race for the Galaxy', a digital adaptation of the board game, uses temporal difference learning to power its AI. This knowledge-free system requires no human input to generate training data, which allows it to improve by playing against itself. Through this approach, the Temple Gate Games team was able to dramatically improve the challenge level offered by AI opponents without the significant time investment typical of tuning complex AI.

Takeaway

Attendees will learn why to include a neural-network in their board game AI, and what kind of games are the best candidates. They'll find out how to maintain performance, architect bifurcations to specialize without sacrificing simplicity and atomity, and auto-generate training data based on a knowledge-free system via temporal difference.