A bias is an adjustable number added to a neuron’s weighted sum. It shifts that sum up or down. Each neuron has its own bias, shared across all examples.
Weighted sum 10 + bias (−3) = 7
The nine red weights multiply signals; the four blue biases adjust the sums. Hover to see a value. Drag through 1,000 positions from −10 to +10, or click to enter a decimal. Values display two decimal places; calculations retain full precision.
Every weight and bias starts independently and uniformly between −10 and +10. Different starts can lead to different results.
One Step adjusts all knobs using all three examples. Train to Completion repeats until improvement becomes negligible or a step limit is reached. A perfect fit is not guaranteed.
Try a New Input predicts without training. Fitting three examples does not guarantee reliable new predictions. Outputs can fall outside 0–100.