This course focuses on data augmentation for image-based reinforcement learning. The learning outcomes and goals include understanding model-free reinforcement learning algorithms for visual continuous control and utilizing data augmentation to learn directly from pixels. The course teaches skills such as implementing image augmentation techniques, exploring different augmentation strategies, adjusting hyperparameters, and pre-training task-agnostic representations. The teaching method involves a seminar format with a presentation divided into two parts. The intended audience for this course includes individuals interested in reinforcement learning, computer vision, and artificial intelligence research.
Data Augmentation for Image-Based Reinforcement Learning
Massachusetts Institute of Technology via YouTube
Overview
Syllabus
Introduction
Outline
Problem
Image Augmentation
Other Augmentation Strategies
Hyper Parameters
Models and Auxiliary Tasks
Results
Atari Benchmark
Image Augmentations
Summary
Dr Q
Dr Qv2
Dreamer
Conclusion
Reinforcement with prototypical representations
Limitations
Task Exploration
Selfsupervised Learning
ProtoRL Approach
Example
Importance of Exploration
Benchmarking
Wrapup
Taught by
MIT Embodied Intelligence
