Batch Size

Friendly Description: Batch size is just how many examples an AI looks at before it pauses to learn a lesson. Imagine a teacher grading homework: she could grade one paper, then update her teaching, or she could grade thirty papers and then notice the patterns. Bigger batches give a smoother view of how the class is doing, but smaller batches let her react more quickly to surprises.

Example: When training a model on photos, a developer might pick a batch size of 32. That means the model looks at 32 photos, calculates how wrong its guesses were on average, and then nudges its internal settings before moving on to the next 32 photos. They keep doing this until the model has seen everything.