This slide introduces the model which is one of the deep Q network. Dueling Network is the successor model of DQN or DDQN. You can easily understand the architecture of Dueling Network.
2. 今回取り上げるのはこれ
[1] Z. Wang, et. al “Dueling Network Architectures for Deep
Reinforcement Learning.”
arXiv1511.06581. 2016.
Q値をV値と行動aに分離することにより性能を向上させ
た!
6. まず強化学習の基本から
the value of the state-action Qπ
s,a( )= E Rt st = s,at = a,π⎡⎣ ⎤⎦
Vπ
s( )= E
a≈π a( )
Qπ
s,a( )⎡⎣ ⎤⎦the value of the state
st
st+1 st+2
st+2st+1
st+1
at
1
at
2
at
3
Qπ
s,a( )
Vπ
s( )
7. the advantage functionを定義
the value of the state-action Qπ
s,a( )= E Rt st = s,at = a,π⎡⎣ ⎤⎦
Vπ
s( )= E
a≈π a( )
Qπ
s,a( )⎡⎣ ⎤⎦the value of the state
st
st+1 st+2
st+2st+1
st+1
at
1
at
2
at
3
Qπ
s,a( )
Aπ
s,a( )= Qπ
s,a( )−Vπ
s( )the advantage function
Vπ
s( )
差をとってる
から を引いて とする Vπ
Qπ
Aπ