How to under the tensor
tensor
what is tensor ?
We talk about the ” Tenser” which is used in ML community.
Actually, the tenser is used as a talk and terminology to refine and rename things that already exist:
Call the “scalar” (a data) as the zero dimensional tensor;
Call the “array” in the programming field as the one dimensional tensor;
Call the “matrix” as the two dimensional tensor;
Call the “video” as the n dimensional tensor;
So, it is boring from the perspective of the gay who create the definition of tensor.
what is the big deal ?
It holds data, weights and biases.
Can realize the hardware acceleration and Autograd !
how to understond the relative concept about the tensor ?
The chain rule:
The Chain Rule is a fundamental concept in calculus. It provides a method for finding the derivative of a composite function that is composed of two or more functions.
Backpropogration:
It is the application of the chain rule, It is an efficient method for computing the gradient of the loss function with respect to theweights of the network, enabling the adjustment of weights to minimize the error.
The automatic differentiation:
is the automatic tool to realize the backpropogation. Of course, the forepropogation also can be realized by the autograd of the Pytorch.
reference:
backpropogation
tensor——statQuest
chain rule