Skip to content

PointerRain/DesmosDigitClassifier

Repository files navigation

I trapped an AI in a calculator.

Here's how I did it

The graph

https://www.desmos.com/calculator/i9pwenswgj

How it works

The model consists of one convolutional layer with a 5x5x4 kernel, two max pooling layers, three fully connected layers, and a softmax output layer. The input is a 28x28 image of a handwritten digit, and the output is a prediction of which digit it is. Each layer is implemented in Desmos using a combination of functions and lists. The convolutional layer uses a kernel to convolve the input image, the max pooling layers downsample the feature maps, and the fully connected layers compute the final prediction. The model is trained using the MNIST dataset in Python Keras.

That's a 4
Yes, yes it is.

Convolution of a 4
This looks less like a 4. (But it still is!)

Progress Updates

Able to draw numbers in the graph. Better hope it's a nine because that's all it ever says.

Added a basic kernel system. Also added draw and clear buttons.

Vastly improved the drawing system with antialiasing.

Convolutions system and MaxPool2D.

Added all the layers, relu, and softmax. Now it thinks everything is a seven though.

Fixed all the bugs and added a prediction display.

Clean up and optimisation.

Optimisations and aligning the output display.

References

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages