University of Twente Student Theses

Login

Generative replay in deep reinforcement learning

Ludjen, A.C.P.P. (2021) Generative replay in deep reinforcement learning.

[img] PDF
543kB
Abstract:Deep reinforcement learning has been an improvement for reinforcement learning, in which it allows the learning of much more complicated environments. However, the issue of catastrophic forgetting is present in deep reinforcement learning. Catastrophic forgetting is a situation where the agent of deep reinforcement learning forgets past experiences. Experience replay is introduced following this issue, by allocating a buffer to store past experiences and to sample from them in the process of learning. A notable downside of experience replay lies in memory usage. Having the possibility of consuming a huge memory space, reduces the scalability of deep reinforcement learning. This research explores the alternative of experience replay in the form of generative replay. Generative adversarial network, a generative model, is implemented in combination with deep reinforcement learning to assess its performance in comparison to experience replay.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:http://purl.utwente.nl/essays/87315
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page