mathematics//probability

Probability is the branch of mathematics that keeps the books of uncertainty: it gives every possible outcome a number between zero and one saying how credible it is, with rules for combining those numbers, and it is used wherever a system measures or decides without knowing everything. The bearing temperature ten minutes from now, the reading of a drone's gyroscope at rest, the day the next pump fails: none of them needs the world to be random in any deep sense. It is enough that the engineer does not know, and probability is that ignorance written with decimals.

Probability is the branch of mathematics that keeps the books of uncertainty: it gives every possible outcome a number between zero and one saying how credible it is, with rules for combining those numbers, and it is used wherever a system measures or decides without knowing everything. The bearing temperature ten minutes from now, the reading of a drone's gyroscope at rest, the day the next pump fails: none of them needs the world to be random in any deep sense. It is enough that the engineer does not know, and probability is that ignorance written with decimals.

The vocabulary starts with the random variable, a quantity not known exactly, and its probability distribution, how much credibility each value carries. The distribution is rarely known whole, so it is summarized by its expected value and its variance, which for a sensor error are the bias that calibration removes and the noise that averaging shrinks, and for two variables by their covariance. Bayes' rule updates a distribution when a reading arrives, and every estimator from the Kalman filter to the particle filter applies it once per cycle.

Every number a system produces carries a doubt, and the doubt has a shape.

Carrying the shape along (a mean and a covariance at least) is what lets a filter weigh two sensors, a threshold trade false alarms against misses, and a planner put margins where the uncertainty is.

When the doubt unfolds in time, the model is a stochastic process: the white noise of a sensor, the random walk of an integrated gyroscope, the slow wander of a bias, the Markov chain of a machine moving between healthy, degraded and failed, with its long run given by the stationary distribution.

When the doubt is about waiting, robots at a charger, messages on a bus, tasks on a processor, the model is queueing theory, whose main lesson is that random arrivals make waits explode long before a server is full.

When the question is which action to take, probabilities are joined to costs in decision theory; when the probabilities themselves have to be learned from data, the reasoning runs the other way, from observed outcomes back to the process that produced them, and that is statistics.