Predicting handwritten digits

Copyright (c) 2020 Michio Inoue

This is a practice mini-project to learn HTML, JavaScript, and WebAssembly. 日本語解説はこちら

MATLAB's classical neural net is implemented as a form of WebAssembly, through "Generate JavaScript Using MATLAB Coder" version 2.0.2 by Geoff McVittie. The tool allows you to create JavaScript/WebAssembly libraries from MATLAB projects using MATLAB Coder.

You can find the tool here: https://jp.mathworks.com/matlabcentral/fileexchange/69973-generate-javascript-using-matlab-coder

• Step 1: Draw a digit
• Step 2: Click [Prediction]

Please note that the current version produces very low accurate results.

The HTML is based on https://github.com/yukagil/tfjs-mnist-cnn (Thank you!)

ToDo

1. Documentation.
2. Update the predictive model. The current network has only one hidden layer and trained with 500 digit images only.
3. Try other algorithms (eg. HOG + SVM).