Interactive MNIST Digit Classifier

Draw a digit and watch a neural network classify it in real-time

Created: January 28, 2025 | Categories: ai-ml

🚀 Launch Full Demo 📁 View Source Code

Live Demo

Note: In a real implementation, you would embed your interactive demo here

Interactive MNIST Classifier Demo Would Go Here

Draw a digit and see the neural network's predictions!

How It Works

This demo uses a convolutional neural network trained on the MNIST dataset, running entirely in your browser using TensorFlow.js.

Model Architecture

const model = tf.sequential({
  layers: [
    tf.layers.conv2d({
      inputShape: [28, 28, 1],
      kernelSize: 3,
      filters: 32,
      activation: 'relu'
    }),
    tf.layers.maxPooling2d({poolSize: 2}),
    tf.layers.conv2d({
      kernelSize: 3,
      filters: 64,
      activation: 'relu'
    }),
    tf.layers.maxPooling2d({poolSize: 2}),
    tf.layers.flatten(),
    tf.layers.dense({units: 128, activation: 'relu'}),
    tf.layers.dense({units: 10, activation: 'softmax'})
  ]
});

Training Results

• Training accuracy: 99.2%
• Test accuracy: 98.7%
• Model size: 1.2MB

Try drawing different styles of digits to see how the model performs!