Count-Based Exploration via Embedded State Space for Deep Reinforcement Learning

Table 2

Downsampling experiments in Atari 2600. represents the dimension of downsampling. Boldface numbers indicate best results.

Freeway	25.1	34	28.3
Venture	479	677	983
Montezuma	24	74	196
Frostbite	3043	5537	4983