perceptron-asm

Implementation of a single-layer perceptron in x86 assembly.

I was inspired by the very simple perceptron algorithm described in Veritasium's video on analog computers. I also wanted another assembly challenge and I've never written any x86 floating point code before.

Results

This perceptron is a classifier to distinguish between circles and rectangles.

Model trained with 500 samples and 3000 training rounds resulted in 90% success rate.

See model.ppm for the raw image file.
I also generated a video of the training process docs/training.mp4

Perceptron Summary

A perceptron is a simple mathematical model attempting to mimic how a biological neuron works. A neuron fires or activates when the dot product of inputs and weights is larger than the specified bias.

Frank Rosenblatt built the first implementation of a perceptron as a specialized machine in 1958. Read more about the history of the perceptron here.

A single-layer perceptron is the simplest neural network you can make.

Run Locally

• dependencies: apt install nasm ffmpeg imagemagick
• build: make
• build and run: make run
• build assets: make assets

References

• x86 64-bit Linux System Calls
• Feedforward Neural Network
• Perceptron Wikipedia
• Compiler Explorer
• Veritasium - Future Computers Will Be Radically Different
• Linear Congruential Generator (Easy random numbers in ASM)
• PPM files
  • https://manpages.ubuntu.com/manpages/bionic/man5/ppm.5.html
  • https://people.cs.clemson.edu/~dhouse/courses/405/notes/ppm-files.pdf
  • Netpbm
• Floating point
  • https://en.wikibooks.org/wiki/X86_Assembly/Floating_Point
  • http://www.ray.masmcode.com/tutorial/index.html
  • IEEE-754 Floating Point Converter
  • https://www.cs.cornell.edu/~tomf/notes/cps104/floating.html
  • http://mathcenter.oxford.emory.edu/site/cs170/ieee754/
  • The Art of Assembly Language. Randall Hyde - Chapter 14