https://www.youtube.com/watch?v=VMj-3S1tku0&list=PLAqhIrjkxbuWI23v9cThsA9GvCAUhRvKZ

This seems dense at first, but if you stick with it, it makes sense

Grad and Backpropagation

• In the example in the video - 2 values can be manipulated in many ways to arrive at a final value. The gradient will tell how a change in one of those two values will affect the final value - negative or positive and by how much.
• Grad is done by determining the derivative of the function and then determining the slope
• micrograd is a library that the creator of this video created to automatically grad operations.
  • These values can then be plugged into a tensor - to create more efficiencies.

What are Derivatives?

The goal of a derivative is to get the slope of two functions.

The slope attempts to explain how the result of a function will be affected by a change in one of the values in the function.

$slope = (d2-d1)/h$

d1 is the original function

d2 is the function with the values modified ie. addition or subtraction of one of the values