machine learning - Why do we use gradient descent in linear regression? -

machine learning - Why do we use gradient descent in linear regression? -

in machine learning classes took recently, i've covered gradient descent find best fit line linear regression.

in statistics classes, have learnt can compute line using statistic analysis, using mean , standard deviation - this page covers approach in detail. why seemingly more simple technique not used in machine learning?

my question is, gradient descent preferred method fitting linear models? if so, why? or did professor utilize gradient descent in simpler setting introduce class technique?

the illustration gave one-dimensional, not case in machine learning, have multiple input features. in case, need invert matrix utilize simple approach, can hard or ill-conditioned.

usually problem formulated to the lowest degree square problem, easier. there standard to the lowest degree square solvers used instead of gradient descent (and are). if number of info points hight, using standard to the lowest degree squares solver might expensive, , (stochastic) gradient descent might give solution in terms of test-set error more precise solution, run-time orders of magnitude smaller (see great chapter leon bottou)

if problem little can efficiently solved off-the-shelf to the lowest degree squares solver, should not gradient descent.

machine-learning statistics linear-regression