DL implementation study, Jan 13, 2024

Building micrograd

Notes

  • [DONE] The intuition behind derivative of a multivariate function
  • A NN can be represented purely mathematically or as a graph data structure in a programming language. Remember, at the end of the day, they are just languages and we can use any language to represent or describe the construct, in this case a NN, we have in our head.

Questions

Why are gradient descent or its variants used to optimize parameters instead of solving the loss function analytically for its minima?

AN: complex loss functions aren’t easily solvable analytically

Leave a comment