Basically, in various machine learning algorithms we need to minimize the error and in a few cases the shape of the error function is concave which is bowl shaped.If the error function is quatric then we can expect to find a closed form solution like in linear regression but in some cases closed form solution is not possible.In that case, we can use stochastic gradient descent.In this, we will start from a random point and will move in direction opposite to the derivate until we reach the minima.In stochastic, instaed of taking the entire input matrix at once, we take one input at a time,f ind the error and then subtract th derivative of the error function with respect to the weight from the weight.