Abstract
Driven by the recent technological advancements within the field of artificial intelligence research, deep learning has emerged as a promising representation learning technique across all of the machine learning classes, especially within the reinforcement learning arena. This new direction has given rise to the evolution of a new technological domain named deep reinforcement learning, which combines the representational learning power of deep learning with existing reinforcement learning methods. Undoubtedly, the inception of deep reinforcement learning has played a vital role in optimizing the performance of reinforcement learning-based intelligent agents with model-free based approaches. Although these methods could improve the performance of agents to a greater extent, they were mainly limited to systems that adopted reinforcement learning algorithms focused on learning a single task. At the same moment, the aforementioned approach was found to be relatively data-inefficient, particularly when reinforcement learning agents needed to interact with more complex and rich data environments. This is primarily due to the limited applicability of deep reinforcement learning algorithms to many scenarios across related tasks from the same environment. The objective of this paper is to survey the research challenges associated with multi-tasking within the deep reinforcement arena and present the state-of-the-art approaches by comparing and contrasting recent solutions, namely DISTRAL (DIStill & TRAnsfer Learning), IMPALA(Importance Weighted Actor-Learner Architecture) and PopArt that aim to address core challenges such as scalability, distraction dilemma, partial observability, catastrophic forgetting and negative knowledge transfer.
Highlights
Reinforcement learning (RL) has established its position as a vital technology in domains such as robotics and intelligent agents [1]
One major challenge concerning multi-tasking within deep reinforcement learning is related to establishing a balance between the needs of multiple tasks within the environment competing for the limited resources of a single learning system
This literature review was conducted with the objective of surveying and analyzing various methodologies that are developed for the optimization of the reinforcement learning (RL) agent’s multi-tasking learning capabilities with the help of deep reinforcement learning
Summary
Reinforcement learning (RL) has established its position as a vital technology in domains such as robotics and intelligent agents [1]. As part of the survey efforts, multiple literature survey papers were examined that are predominantly related to the foundations of deep reinforcement learning and its applicability One of these surveys analyzed the foundations of reinforcement learning, which covers core elements such as dynamic programming, temporal difference learning, exploration vs exploitation, function approximation, and policy optimization [11]. The survey focus was directed toward addressing multiple research challenges associated with the application of multi-tasking in deep reinforcement learning; it examined the three major state-of-the-art solutions that are implemented to overcome some of those challenges. Throughout this literature survey, the key focus remained on investigating various methodologies that are related to multi-tasking-related aspects.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have