Machine learning is changing the way we make games, and the 2019 Game Developers Conference is the place to be if you're looking to better understand how machine learning techniques can be applied across different disciplines to help you build great games!
The first thing you should look at is the GDC 2019 Machine Learning Tutorial, a special day-long deep dive into the state of machine learning in game development, and what sorts of tricks and techniques are being used by the best in the business!
In his Machine Learning Tutorial talk on "Beating Wallhacks using Deep Learning with Limited Resources", for example, Nexon Korea's Junsik Hwang will show you how Nexon Korea has developed a real-time automated wallhack detection system using Convolutional Neural Networks with a small dataset and a single GPU.
By using Class Activation Maps, the network finds suspicious areas within a screenshot that improves the credibility of the model's performance and makes debugging datasets much more efficient. Model Interpretability plays a crucial role in incorporating deep learning with the existing abuser control policies. As a result, the system now detects abusers in real-time and reduces manual inspection labor significantly!
Plus, there's a smorgasbord of intriguing GDC 2019 talks outside the Machine Learning Tutorial that present different perspectives on the best ways to use ML in your own games. Among them is "A New Era of Performance Capture with Machine Learning", a session presented by Ubisoft Montreal's Daniel Holden that's all about how ML tech is changing the way dev teams handle performance capture.
In his talk Holden will dig into the details of several new performance capture technologies developed at Ubisoft La Forge, including a neural network used to automatically clean motion capture data, an end-to-end solution for capture of facial animation from video, and a machine learning solution which can generate facial animation from raw audio. Don't miss out!
"Building Abusive Chat Detection Systems with Deep Learning" is another promising GDC 2019 talk you'll want to see if you're at all curious about this stuff. Presented by Blizzard senior data scientist Ryan Brackney, this talk aims to provide you with a technical framework for building an abusive chat detection system with deep learning.
Any system for moderating the communication of an in-game, global community of players will face a number of challenges including how to handle the unique and ever-changing lingo of players, accommodate multiple languages and cultural values simultaneously, and being robust against players trying to circumvent the detection methods. This talk will describe in detail the deep learning architecture and systems level solutions Blizzard chose to tackle these problems, with an eye towards helping you apply what Blizzard learned to your own solutions.
Audio experts are encouraged to check out "Introduction to Machine Learning for Game Audio", because speaker John Byrd (CEO of Gigantic Software) will give you a quick summation of the most relevant research in this exciting field, and describes novel applications of machine learning, specifically for interactive audio.
You'll see demonstrations and code for: automatic feature extraction, classification, speaker detection, emotion detection, and character synthesis. And, you'll get pointers to papers and open source implementations of these algorithms. Expect to walk away with a list of rules for what will and what will not be possible with deep learning, until the year 2030, and some new ideas for how to apply it to your own projects!
Further details on these machine learning talks and many more are available now on the GDC 2019 Session Scheduler. There you can begin to lay out your GDC 2019, which takes place March 18th through the 22nd at the (newly renovated!) Moscone Center in San Francisco.
Bring your team to GDC! Register a group of 10 or more and save 10 percent on conference passes. Learn more here.
For more details on GDC 2019 visit the show's official website, or subscribe to regular updates via Facebook, Twitter, or RSS.
