A4.3.8 HL · neural networks

The perceptron

A perceptron is one artificial neuron: it draws a single straight line to split its inputs into two. Train it on a logic gate and watch the line rotate into place. Then try XOR, and meet the limit that a single neuron cannot pass.

Each dot is one truth-table row, placed by its two inputs; the dotted square joins the four input pairs (blue = output 1, orange = output 0). The line is the perceptron's boundary; a dashed ring marks a row it still gets wrong.

Gate to learn

the truth table, live

in 1	in 2	AND	model
0	0	0	0 ✓
0	1	0	0 ✓
1	0	0	0 ✓
1	1	1	0 ✗

The model column is what the current line answers for each row; training is done when it matches the AND column on every row.

epoch

rows wrong

Each epoch nudges the line to fix a wrong row. AND, OR and NAND are one line apart. XOR is not, and the line will keep wobbling forever.

loading…

← Machine learning

A4.3.8 HL · neural networks

The perceptron

Gate to learn

the truth table, live

in 1	in 2	AND	model
0	0	0	0 ✓
0	1	0	0 ✓
1	0	0	0 ✓
1	1	1	0 ✗

The model column is what the current line answers for each row; training is done when it matches the AND column on every row.

epoch

rows wrong

Each epoch nudges the line to fix a wrong row. AND, OR and NAND are one line apart. XOR is not, and the line will keep wobbling forever.