Skip to content

This project is part of a 3-year project funded by the National Science Foundation to bring AI Education to underrepresented high schools in Tennessee. The project is done by Professor Pengyu Hong (Brandeis University) and his research group. We hope to use this visualization tool to spark interests in coding and linear algebra within those student

License

Notifications You must be signed in to change notification settings

MARCOpo1o/visualize_neural_net

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

An Interactive Node-Link Visualization of Convolutional Neural Networks

Visualization

Introduction

Project ID: NSF DRL-2314156 This project showcases what each layer of neurons is highlighting in a convolutional neural network (CNN). The aim is to display the effects of deeper feature map layers on the output.

Motivation

Sponsored by the National Science Foundation (NSF), this project is part of a 3-year initiative to bring AI education to underrepresented high schools in Tennessee. The project is done by Professor Pengyu Hong (Brandeis University) and his research group. We hope to use this visualization tool to spark interests in coding and linear algebra within those students. I work on the visualization of Convolutional Neural Network and will be used for teaching.

How to Use

Visit the live demo and draw a digit on the canvas. The network will then show you the activation of each layer of neurons.

Features and Updates

1. Full Layer Connections:

Hovering over a node shows connections from that node back to the input layer, displaying the full path through the network, providing a clearer understanding of how features are propagated through the network. Full Layer Connections

2. Randomize Weights:

Added buttons to randomize weights in the two convolution layers. This allows users to explore the effect of pre-trained weights on outputs. Users can also make all weights identical and revert changes. This is applied to both convolution layers. The following example is for the first convolution layer. weight_buttons

Original Weights

Original Weights

Identical Weights

Identical Weights

Randomized Weights

Randomize Weights

3. Enhanced Output Layer:

Enlarged and labeled the output layer for better visibility. output

4. Fully Connected Layers:

Edges can switch between displaying hidden weights and the output from the previous layer. button show weight fully connected layer showing hidden weights button show node output fully connected layer showing node output

5. Overall Network Structure:

Added an option to show an image of the entire CNN network structure used in this visualization. network structure

Acknowledgments

This project builds on code by Adam Harley. I would like to thank Professor Pengyu Hong from Brandeis University for his continuous mentorship and guidance, Zhengyang Zhou (PhD candidate at Brandeis University), and other members of Professor Hong's group for their help in testing the website. Special thanks to Kanh (Andrew) Nguyen for introducing me to this opportunity.

Citation

A. W. Harley, "An Interactive Node-Link Visualization of Convolutional Neural Networks," in ISVC, pages 867-877, 2015

About

This project is part of a 3-year project funded by the National Science Foundation to bring AI Education to underrepresented high schools in Tennessee. The project is done by Professor Pengyu Hong (Brandeis University) and his research group. We hope to use this visualization tool to spark interests in coding and linear algebra within those student

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • JavaScript 56.3%
  • HTML 35.5%
  • CSS 8.2%