Start here

A school's water tank has a gauge. The gauge shows how many steps the water is below full. The caretaker reads it. Then she waits, or calls the water tanker. A tanker is a truck that brings water. This track stores her rule as a few numbers. It does not keep a pile of past days.

What you already own

• Labelled examples. Each example day has a word, call or wait.
• The kept-back count. Kept-back days are not learned from. They test the rule. Accuracy is a count with its denominator. A denominator is the total a count is out of. It returns here.
• The nearest-day rule. It keeps every example. The rule in this track keeps almost nothing.

What this track covers

• Holding a rule instead — two weights and a threshold, and nothing else.
• The weighted sum — each input times its weight, then added.
• The line it crosses — one number the sum must meet.
• Measuring one wrong — the kept-back count finds a miss.
• Nudging the weights — one learning step worked by hand.
• Counting what moved — the count, done again.
• Write the sum yourself — two short functions.
• What the word neural buys — what a borrowed word does not claim.

This is How AI Learns II: Neural Networks Intuition. A caretaker's rule is not a rule of numbers yet. That is what this track changes.