Learning to Act by Watching Unlabeled Online Videos - Paper Explained
Offered By: Yannic Kilcher via YouTube
Course Description
Overview
Explore a comprehensive analysis of OpenAI's Video PreTraining (VPT) technique for tackling the complex challenge of creating an AI agent capable of playing Minecraft. Delve into the innovative approach of leveraging a small set of labeled contractor data to pseudo-label a vast corpus of scraped gameplay footage. Discover the model architecture, experimental results, and fine-tuning processes that led to the first Minecraft agent achieving the impressive feat of crafting a diamond pickaxe autonomously. Learn about the semi-supervised imitation learning method used to extend internet-scale pretraining to sequential decision domains, and understand how this behavioral prior demonstrates nontrivial zero-shot capabilities. Gain insights into the potential applications of this technology in robotics, video games, and computer use, as well as the hardware considerations for implementing such advanced AI systems.
Syllabus
- Intro
- How to spend money most effectively?
- Getting a large dataset with labels
- Model architecture
- Experimental results and fine-tuning
- Reinforcement Learning to the Diamond Pickaxe
- Final comments and hardware
Taught by
Yannic Kilcher
Related Courses
Computational NeuroscienceUniversity of Washington via Coursera Reinforcement Learning
Brown University via Udacity Reinforcement Learning
Indian Institute of Technology Madras via Swayam FA17: Machine Learning
Georgia Institute of Technology via edX Introduction to Reinforcement Learning
Higher School of Economics via Coursera