S E E D / / S E A R C H F O R E X T R A O R D I N A R Y E X - PowerPoint PPT Presentation

S E E D / / S E A R C H F O R E X T R A O R D I N A R Y E X P E R I E N C E S D I V I S I O N

Deep Learning in Games Martin Singh-Blom @singhblom

S E E D // Introduction What is Machine Learning?

S E E D // Introduction What is Artificial Intelligence?

S E E D // Introduction or ...

S E E D // Introduction Guess the function!

S E E D // Introduction f (2) = 4

S E E D // Introduction f (8) = 16

S E E D // Introduction f ( x ) = 2 x

S E E D // Middle How does the machine guess?

S E E D // Middle It learns from the data.

S E E D // Middle It learns from the data. (That’ s why we call it machine learning!)

S E E D // How do the machines learn? Guess a straight line! y x

S E E D // How do the machines learn? Guess a straight line! y f ( x ) = 9.5 – 0.3 x x

S E E D // How do the machines learn? Guess a straight line! f ( x ) = 0.3 + 0.6 x y x

S E E D // How do the machines learn? Guess a straight line! f ( x ) = 0.3 + 0.6 x f ( x ) = 9.5 – 0.3 x

S E E D // How do the machines learn? That is all there is to it! 1. Data – f ( x ) = y pairs. 2. A way to tell the machine how bad a guess is. 3. Some idea of what kind of function the machine is allowed to guess – straight line? Curve? Something stranger?

S E E D // How do the machines learn? Guess a straight line! y x

S E E D // How do the machines learn? Guess a straight line!

S E E D // Deep Learning What is Deep Learning?

S E E D // Deep Learning What are Artificial Neural Networks?

S E E D // Deep learning f

S E E D // Deep Learning

S E E D // Deep learning

S E E D // Deep learning f ( ) = 8

S E E D // Deep learning f ( )

S E E D // Deep learning f ( ) = cat

S E E D // Deep learning ”I saw it in a theater once and it was great. It was very… I don’t know, f ( ) = ” Have you seen a little dark. Suicide Squad ?” I like the psychological effects and the way it portrays the characters.”

S E E D // Deep learning f ( ) = ”A person flying a kite on a beach”

S E E D // Deep learning f ( ) = ”A coffee, please.”

S E E D // Deep learning f ( ) =

S E E D // Agents in Games Agents in Games

S E E D // Agents in Games

S E E D // Agents in Games f ( ) =

S E E D // TOPIC

AlphaGo

S E E D // Animation Animation

S E E D // Animation

S E E D // Animation f ( ) =

Audio-Driven Facial Animation by Joint End-to-End Learning of Pose and Emotion, Karras et al., 2017, NVIDIA

S E E D // All the things! Learn all the things!

S E E D // All the things! f ( ) =

Physics Physics Forests: Real-time Fluid Simulation using Machine Learning, Ladicky et al., 2015, www.physicsforests.com

S E E D // All the things!

Realtime Multi-Person 2D Human Pose Estimation using Part Affinity Fields, Cao et al., 2017

Phase-Functioned Neural Networks for Character Control, Holden, 2017

S E E D // All the things! f ( ) = f ( ) = f ( ) = cat ”I saw it in a theater once and f ( ) = it was great. It was very… I don’t know, ” Have you seen f ( ) = Suicide Squad ?” a little dark. I like the psychological effects and the way it f ( ) = 8 portrays the characters.” f ( ) = f ( ) = ”A coffee, please.” f ( ) = ”A coffee, please.”

S E E D // Final remark It’s all just function guessing – or – A new paradigm for computing Instead of programming – showing Same method for every problem Greatest paradigm change in computing since transistors

S E E D // Thank you Stockholm Los Angeles Mark Kyobe Effeli Holst Hector Anadon Leon Carlos Ochoa Jenna Frisk Jorge del Val Santos JP Lewis Ida Winterhaven Mattias Teye Binh Le Tomasz Stachowiak Anastasia Opara Henrik Halen Colin Barré -Brisebois Camilo Gordillo John Courte Graham Wihlidal Joakim Bergdahl Lars Sjöström Jack Harmer Daniel Lundin Special thank you to Linus Gisslén Henrik Johansson Magnus Nordin Montreal Paul Greveson Johan Andersson Niklas Nummelin Mathieu Lamarre Ken Brown Etienne Danvoye