1. Notation Explaination:
Put hat on top of a notation means that the new notation represents an estimation of the orginal one.
2. The union bound:
The probability of one of N events happens is less than the sum of the probability of each event happens.
3. The uniform convergence holds true regardless of the constitution of the hypothesis class, which means even if hypothesis in the hy class are from different algorithm type (SVM and BP network) it still applies.
4. Sample Complexity Bound
How large the training set should be in order to achieve the certain bound on the errors ?