A small educational React website to understand how backpropagation and chain rule work in neural networks.
Classic calculus problem
2-1 fully connected neural network:
Interactive Computational Graph
- Show how derivatives are calculated in any scalar conputational graph
- Customizable operators
- Can save/load the graph to/from file
For simplicity, the built-in operators and how derivatives are accumulated are limited to scalar values, but the architectural design is open to handle higher-dimensional values.
VSCode can be used to develop the website.
If you use VSCode, please open the subdirectory ./interactive-computational-graph
in a new Window and enable the extensions like ESLint in the new window. Otherwise, we have to set up some extra settings if we open a VSCode window here.
- Jest: Runs Jest tests
- Code Spell Checker: Checks common spelling errors
- ESLint: Lints Typescript files
- Stylelint: Lints CSS files
- Prettier: Formats Typescript files
- Markdown Preview Mermaid Support: Previews Mermaid graphs in Markdown files