gradient descent
nounUK /ˈɡreɪdiənt dɪˈsent/
US /ˈɡreɪdiənt dɪˈsent/
A way a computer learns by taking small steps to make its mistakes smaller and smaller.
The programmer used gradient descent to help the model make better predictions.
OTHER MEANINGS
2. An optimization algorithm that repeatedly moves in the direction that reduces a function fastest.
Gradient descent found the lowest point of the cost function after many iterations.
3. A training technique in neural networks where weights are adjusted step by step to lower error.
Stochastic gradient descent speeds up training by updating weights on small batches.
4. A general strategy of improving by following the steepest downward slope of a problem.
Think of gradient descent as walking downhill in fog to reach the valley.