Uploaded August 2022 | Updated September 2026, 3 weeks ago

---

A demo is should be available at graemeniedermayer.github.io/ArExperiments/NNVis.html

This is a visualization of the wieghts of AlexNet a neural network that won a competition in 2012 that kicked off some of the more recent advancements in artificial intelligence. Many more recent networks are much more complicated but they are still often represented by blocked of weights/parameters.

Time Codes
0:00 AlexNet
1:17 Basic Walkthrough
2:30 More Through Walkthrough
8:04 Image To Weights
9:01 Visualizing with Units or Connections
10:06 Code

There's a lot of sources
medium.com/@sdoshi579/convolutional-neural-network-learn-and-apply-3dac9acfe2b6
github.com/mikechen66/AlexNet_TensorFlow2.0-2.3 (this was the alexnet version I used)
en.wikipedia.org/wiki/AlexNet
proceedings.neurips.cc/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf (alexnet paper)

This visualisation contains all of the 60,954,656 weights of AlexNet. Gosh I need to learn how to do 3d video stabilization.

Code is at
github.com/graemeniedermayer/ArExperiments/blob/main/javascript/arVisNN.js