What is an Epoch?
In machine learning, one epoch means that the training algorithm has made exactly one complete pass through all the available examples in the training dataset.
How does it work?
A model rarely learns everything it needs to know by looking at the data just once. Training commonly requires multiple epochs. After the first epoch, the model reviews the same dataset again for a second epoch, slowly refining its understanding and improving its accuracy with each complete pass.
What is a common misconception?
More epochs are not automatically better. If a developer forces the model to run through too many epochs, the model will stop learning general patterns and start memorizing the specific training examples—a failure state known as overfitting.
Why does it matter?
The number of epochs is a fundamental hyperparameter that dictates the computational cost and time required to train an AI. Balancing the number of epochs to achieve high accuracy without overfitting is a core challenge in machine learning engineering.