Epoch

Definitions

1. One complete pass through the entire training dataset. Models are typically trained for multiple epochs — enough to learn patterns but not so many that they overfit. Finding the right number of epochs is more art than science.

   In plain English: One full pass through all the training data when teaching an AI model — like reading a textbook cover to cover once. Models usually need several passes to learn well.

2. In computing, 'epoch' also refers to a fixed reference point in time from which time is measured. Unix epoch is January 1, 1970, 00:00:00 UTC — the moment from which Unix timestamps count seconds. The 'Year 2038 problem' occurs when 32-bit Unix timestamps overflow.

   Example: 'The Unix epoch is January 1, 1970. On January 19, 2038, 32-bit timestamps overflow. Hopefully your systems use 64-bit integers by then.'

   Source: Unix epoch / computing

Related Terms

• Fine-Tuning
• Overfitting
• Training Data
• Underfit
• Model Fine-Tuning Cost
• NVIDIA H100
• Garbage In Garbage Out