ideas from statistical mechanics. The advantage of our theory over the
well-established Vapnik-Chervonenkis theory is that our bounds can be
considerably tighter in many cases, and are also more reflective of t
he true behavior of learning curves. This behavior can often exhibit d
ramatic properties such as phase transitions, as well as power law asy
mptotics not explained by the VC theory. The disadvantages of our theo
ry are that its application requires knowledge of the input distributi
on, and it is limited so far to finite cardinality function classes. W
e illustrate our results with many concrete examples of learning curve
bounds derived from our theory.