notesum.ai

Published at December 9

Unraveling the Complexity of Memory in RL Agents: an Approach for Classification and Evaluation

cs.LG

cs.AI

Released Date: December 9, 2024

Authors: Egor Cherepanov¹, Nikita Kachaev¹, Artem Zholus², Alexey K. Kovalev¹, Aleksandr I. Panov¹

Aff.: ¹AIRI, Moscow, Russia; ²Chandar Research Lab

Arxiv: http://arxiv.org/pdf/2412.06531v1

Refer to caption

Hyperparameter	Value
Number of layers	2
Number of attention heads	2
Hidden dimension	256
Batch size	64
Optimizer	Adam
Learning rate	3e-4
Dropout	0.1
Replay buffer size	1e6
Discount ( $\gamma$ )	0.99
Entropy temperature	0.1