Watch a single artificial neuron compute: weights, bias, and one activation function

TinfoilTricorn · x · 2026-11-20

A visual explainer of how little a single artificial neuron actually does: three inputs are scaled by weights, summed with a bias, and passed through one squashing function. The live demo lets you watch activations switch between sigmoid, tanh, and ReLU — with ReLU, anything below zero simply doesn't fire — while every number updates in real time as the weights drift.

Original post →

More from Research

Research channel →