
References
485
learning theory. In fact, Sutton and Barto (1998) indicate that such methods
may be more appropriate than Q-learning and related methods when one faces a
non-Markovian task.
However, such methods are only appropriate for tasks with clearly defined
episodes. The fighter aircraft task is clearly episodic, since engagements take
place in a predefined time window. When one directs an LCS at an episodic task,
epochal credit assignment schemes, like that used in the fighter aircraft LCS,
may be of higher utility than methods based on Q-learning, SARSA, or the
bucket brigade.
Like the Q-learning-based ...