derivative
/ duh-RIV-uh-tiv /
A derivative answers one question: if I nudge the input a tiny bit, how fast does the output change? It is the rate of change, the slope, the instantaneous steepness. When you are driving, your speedometer is showing a derivative — the rate at which your distance is changing right now. Climbing a hill, the derivative is how steep the ground is under your feet at this exact step.
Picture zooming in on a curve until it looks almost like a straight line; the slope of that little line is the derivative at that point. A positive derivative means the output is rising as the input grows; negative means it is falling; and zero means you are momentarily flat — at the top of a hill, the bottom of a valley, or a brief plateau. That last case is precious, because flat spots are where peaks and troughs hide.
Machine learning runs on derivatives. Training a model means tweaking its numbers to shrink its error, and the derivative of the error with respect to each number tells you which way to nudge it and how urgently. That is the whole engine behind gradient descent and backpropagation. The one honest limit: a derivative only describes the immediate neighborhood. It tells you the slope where you stand, not what the landscape looks like a mile away — which is why training proceeds in cautious little steps rather than confident leaps.
For the curve y = x², the derivative is 2x. At x = 3 the slope is 6 — climbing steeply. At x = 0 the slope is 0 — flat, the very bottom of the bowl. The derivative changing from negative to positive as x passes 0 is exactly how calculus locates that lowest point.
The slope of y = x² at any point is 2x; where it hits zero, you've found the bottom of the bowl.
A zero derivative flags a flat spot, but does not by itself tell you whether it is a minimum, a maximum, or just a passing plateau — you need more information to know which. Some functions (like a sharp corner) have no derivative at all at certain points.