/CNNfromScratchCompletely

Scratch Convolutional Network designed as per Lecun et al. 1989, no ML library used whatsoever, only numpy and math :)

Primary LanguagePython

CNNfromScratchCompletely

WORK IN PROGRESS

Scratch Convolutional Network designed as with only numpy and math :)

  • Scratch Archetecture & Custom Backpropogation/Gradient Descent Engine (Like pytorch)
  • Custom Convolutional Neural Network Architecture based on state-of-the-art models
  • No Pytorch or TensorFlow!

Purpose

I've designed this (hopefully) to be as follow-able as possible with plenty of comments and explanations in my code. If anything is unclear, send me a message, and I will try to fix it. With a little work and thinking, you can get a lot out of this rep, as it covers scratch-backpropogation, neural-net archetecture creation, and examining interesting emergent phenomena from (convolutional) neural nets.