Using randomness in a computing device to provide answers. Traditional computers use processors (CPUs, GPUs, TPUs) that are extremely precise, and every effort is made to ensure that all transistors in the circuits open and close exactly when they should. One-hundred percent precision is required every second of every minute. See
CPU,
GPU and
Tensor Processing Unit.
Solve Stochastic Equations
Stochastic equations can solve problems faster for various scientific problems as well as AI. Instead of eliminating noise, drift and randomness in the circuits, which are prerequisites in modern computers, stochastic computers employ those attributes purposefully to achieve results in a much shorter time using much less power.
Percentages of 1s and 0s
Instead of precise binary numbers, the most common stochastic method represents numbers as probabilities. For example, the number 0.80 is represented as a binary string of 80 ones and 20 zeros in a random order. What is important is the ratio of zeros to ones, not their placement in the string.
Stochastic computing demands considerably less power than regular computing, and the accuracy is increased using longer bitstreams. See
Normal Computing,
stochastic parrot and
heuristic.