Iteration
A single weight update — one batch processed and one step taken.
Think of it like
Like a single stir of the pot: one small adjustment, repeated thousands of times to get the dish right.
Example
With 10,000 examples and a batch size of 100, one epoch is 100 iterations.
How it actually works
An iteration (or step) is the atomic unit of training: process one batch, compute the loss and gradient, and update the weights once. The number of iterations equals dataset size divided by batch size, times the number of epochs. Progress and learning-rate schedules are usually tracked in steps, which is why training dashboards count iterations, not epochs.
For product teams
The granular unit training is measured in — loss curves and schedules move step by step.
For engineers
One batch → one gradient update; steps = (dataset/batch) × epochs; the unit for schedules and logging.
Related
- Epoch — Many make up an epoch.
- Stochastic gradient descent — Each does one stochastic gradient descent step.
- Batch size — Sized by the batch size.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome