I'm studying the definition of PAC learnable:
Let C be a concept class over X.
We say that C is PAC learnable if there exists an algorithm A with the following property:
- for every concept c ∈ C,
- for every distribution D on X,
if A is given access to the oracle EX(c, D) and inputs ε, δ, then with probability at least 1 − δ, A outputs a hypothesis concept $c_h$ ∈ C satisfying error($c_h$) ≤ ε,
This probability is taken over the random examples drawn by calls to EX(c, D), and any internal randomization of A.
What is meant by the word concept?