... detectors¹

Non-binary feature detectors providing a discrete range of values can be readily binarized.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... situation.²

Non-Markovian problems, when confronted, should be converted into Markovian ones. How to do that is out of the scope of this paper, although it is one of the most relevant points to achieve a successful real-world reinforcement-learning application.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... holds.³

A partial view can also include negations of feature detectors since the non-detection of a feature can be as relevant as its detection.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.