Introduction - Design and Analysis of Learning Classifier Systems - page 11

an introduction to methods from dynamic programming and reinforcement
learning. Then, the exact role of LCS in handling such tasks is defined, and
a possible method is partially derived from first principles. This derivation
clarifies some of the current issues of how to correctly perform RL with
XCS(F), which is discussed in more detail. Based on the LCS model, it
is also shown how the stability of LCS with RL can be studied, together
with how to handle learning long action sequences and the trade-off between
exploring the space and exploiting current knowledge.
Chapter 10 summarises the work and puts it into the perspective of the initial
objective.