Diafunc's branch of AI

Symbolic AI.

AI you can read.

Symbolic AI is the branch of artificial intelligence that represents knowledge as explicit symbols: mathematical expressions, rules, programs. Not opaque parameters. Where neural networks hide what they learned in millions of weights, Symbolic AI keeps the answer in the open: code you can read, question, and explain.

Symbolic Regression is one technique within Symbolic AI. It searches the space of mathematical expressions for one that explains your data, discovering the formula itself, not just fitting parameters. This is what Diafunc is built on.

Readable doesn't mean small. A symbolic model might be a single equation or a sizeable program. Either way it's written in explicit symbols you can open up, follow step by step, question, and carry somewhere the training data never reached. Because it's out in the open, you can also see where to stop trusting it.

A neural network gives you this

w_1,1 = 0.184,  w_1,2 = -0.273,  w_1,3 = 0.091, …
[2 hidden layers × 128 units × 65,536 parameters]

A Symbolic AI gives you this

f(x) = 0.42 · sin(2π · x₁) + 1.8 · x₂² − 0.31

Same accuracy. Wildly different to live with.

What that buys you

· Every prediction traces back to terms you can read
· A model you can question, correct, and defend, not just accept
· Inspectable whether it's one line or a complex program
· Portable to cases the training data never covered
· Insight into how your system actually works

A central readable formula with mathematical building blocks assembled into the readable model Symbolic AI produces.

The technique behind Diafunc's Symbolic AI is Symbolic Regression.

How it works