Noise robustness experiment 1

[Yoshi-0921]

動的ノイズ(学習)

visible_range = 7

σ={0.3, 0.2, 0.1}	0	σ (68.27%)	2σ (95.45%)	3σ (99.73%)
MAT-DQN	done	done	done	done
Baseline	done	done	done	done

σ={0.07, 0.05, 0.03}	0	σ (68.27%)	2σ (95.45%)	3σ (99.73%)
MAT-DQN	done	done	done	done
Baseline	done	done	done	done

動的ノイズ(評価)

学習環境と評価環境は同じと仮定。ノイズは同様に生成されると仮定。

visible_range = 7

σ={0.3, 0.2, 0.1}	0	σ (68.27%)	2σ (95.45%)	3σ (99.73%)
MAT-DQN
Baseline

σ={0.07, 0.05, 0.03}	0	σ (68.27%)	2σ (95.45%)	3σ (99.73%)
MAT-DQN
Baseline

[Yoshi-0921]

動的ノイズ(評価)

学習環境と評価環境は異なると仮定。ノイズは異なる様に生成される又は生成されないと仮定。

visible_range = 7

σ={0.3, 0.2, 0.1} in σ={0.07, 0.05, 0.03}	0	σ (68.27%)	2σ (95.45%)	3σ (99.73%)
MAT-DQN
Baseline

σ={0.3, 0.2, 0.1} in no noise	0	σ (68.27%)	2σ (95.45%)	3σ (99.73%)
MAT-DQN
Baseline

σ={0.07, 0.05, 0.03} in σ={0.3, 0.2, 0.1}	0	σ (68.27%)	2σ (95.45%)	3σ (99.73%)
MAT-DQN
Baseline

σ={0.07, 0.05, 0.03} in no noise	0	σ (68.27%)	2σ (95.45%)	3σ (99.73%)
MAT-DQN
Baseline

[Yoshi-0921]

ノイズなし(学習)

visible_range = 5

Algorithm	State
MAT-DQN
Baseline

ノイズなし(評価)

Algorithm	State
MAT-DQN
Baseline

[Yoshi-0921]

このノイズの付加方法が実例に基づいていないため、一時休止中