CAP4611 - Fundamentals of Learning (L4.pdf)

overfitting

happens when some boundaries are based on distinctions that don’t make a difference

training data

data given to the model during training; labels are already known

testing data

data for which the trained model will try to predict labels

key principle

“do not let the test data influence the training phase”

IID assumption

examples are sampled independently from the same distribution

independent and identically distributed

IID stands for: ___ ___ ___ ___

learning theory

explores how training error is related to test error

generalization gap

represents the amount of overfitting

fundamental trade-off

“how small can you make the training error” vs. “how close is training error to test error“