AIExplainer

What is an Epoch?

One complete pass through all the training data during model training.

An epoch is one complete pass through all the training data. If a model sees every example once, that counts as one epoch. Training usually requires many epochs before the system performs well.

Too few epochs may underfit; too many without safeguards may overfit.

An epoch is like reading an entire textbook cover to cover once. Re-reading it multiple times — multiple epochs — helps the material sink in, though eventually you stop gaining much from repetition.

Training logs for image models often show loss decreasing over epochs until performance plateaus.

When engineers report that a model was "trained for 50 epochs," they mean it reviewed the entire dataset fifty times — a standard way to describe how thoroughly a model has been trained.

More epochs are not always better — training too long on limited data can hurt generalisation.