Immune-inspired learning algorithm (#altdeeplearning)

Project description: A machine learning algorithm (or family of algorithms) using the immune system as inspiration for adaptation and pattern recognition.

Language: using python (& torch for GPU speedup!)

Hi folks.

I’ve recently been working on an ML algorithm which is based on aspects of the immune system - how it recognises pathogens, how it evolves responses & adapts to new data. It’s currently working on computer vision problems.

The algorithm uses an optimisation technique based on how the immune system creates antibodies, so it doesn’t require back-propagation. It uses cloning and something called “somatic hypermutation” to modify its parameters. It also doesn’t require a gradient to optimise its parameters, so can use functions which don’t have a derivative as affinity functions (similar to DL activation functions).

please let me know if you’re interested, & I’ll post up a writeup/some code!

Best,

Johnny

The idea seems interesting and novel.
I would love to know more about it.