Democratize Machine Learning - ML-Agents Explained

Explore the world of machine learning in game development through this 42-minute conference talk from Unite LA. Discover Unity ML-Agents, an open-source toolkit bridging Unity and machine learning. Learn about recent advances in AI, including reinforcement and imitation learning, and their potential to revolutionize game production. Gain insights into the ML-Agents workflow, understand what agents are and how they can be used in games, and explore various training methods. Follow along with practical examples like the 3D Balance Ball, Multi-Agent Soccer Training, and Curiosity-Driven Exploration. Delve into the challenges of machine learning inference, including computational complexity and platform support, and learn about Unity's innovative solutions. Presented by Arthur Juliani and Vladimir Oster from Unity Technologies, this talk offers a comprehensive overview of how to democratize machine learning in game development.

Intro
Machine Learning for Gaming Al
Unity ML Agents Workflow
What are Agents?
How can agents be used in games?
Set Up Game for Training
Training Methods
Reinforcement Learning Process
Example: Chicken Crossing the Road
Included ML Agents Training Examples
3D Balance Ball
Curriculum Learning
Multi-Agent Soccer Training
Curiosity-Driven Exploration
Pyramids Environment
External reward only
Curiosity and external reward
Machine Learning Inference
Challenges of Inference - Computation Complexity
Challenges of Inference - Platforms to support
Unity Inference Solution
Unity Labs Inference Engine
Thank you!