Review Paper
Year: 2021 | Month: July-September | Volume: 5 | Issue: 3 | Pages: 47-53
DOI: https://doi.org/10.52403/gijash.20210706
How Meta Learning Enhances Reinforcement Learning in AI
Deekshitha Kosaraju
Independent Researcher, Texas, USA
ABSTRACT
Meta learning or the concept of "learning to learn" has become increasingly popular in the realm of intelligence with a focus on reinforcement learning (RL). Traditionally faced with hurdles related to efficiency and adaptability when encountering environments or tasks. However, by incorporating meta learning into RL approaches... Agents are equipped not to grasp individual tasks but also to apply their knowledge to novel tasks, with limited data and exploration. This article delves into the ways in which meta learning boosts reinforcement learning by enhancing adaptability and efficiency in handling samples effectively while also improving generalization capabilities of the system at large. We will delve into the principles of both meta learning and RL and touch upon the hurdles faced by traditional reinforcement learning methods while highlighting how meta learning offers fresh and effective solutions. Furthermore, we will explore the reaching implications of this synergy across various fields such as robotics and healthcare along with autonomous systems which sets the stage for developing AI systems that are more flexible and adept, at various tasks. The article showcases the groundbreaking possibilities of merging meta learning with reinforcement learning in theory and real-world applications using illustrations and real-life scenarios.
Keywords: Meta-learning, Reinforcement Learning, Adaptation, Generalization, Sample Efficiency, Artificial Intelligence, Deep Learning, Robotics, Autonomous Systems, Machine Learning
[PDF Full Text]