The technique behind Ufinq

Symbolic Regression.

Finding the formula in the data.

Symbolic Regression searches the space of mathematical expressions for one that explains your data. Not the parameters of a fixed-form model, but the formula itself: shape, operations, constants, all discovered.

Most modeling picks a shape first and tunes it to fit. Symbolic Regression does the opposite: it lets the shape itself compete. Thousands of candidate formulas rise and fall until the one that best explains the data survives.

Ufinq is Diafunc's Symbolic Regression technology, and it goes beyond evolutionary search: what this page describes is its foundation. The whole method, with a real model and its limits, is in Symbolic Regression, explained.

A typical search

x₁
x₁ · x₂
sin(x₁) + x₂
a · sin(b · x₁) + c · x₂² − d

Generations of competing programs. One wins.

How the search works

· Every candidate is tuned and simplified before it competes
· Decompositional learning splits hard problems apart and recomposes them
· The search runs at scale, across generations of competing programs
· One model comes back, automatically as simple as the data allows
· The research, papers, and benchmarks live at ufinq.com

The model chapter of the Yacht Hydrodynamics report, Formula tab: Residuary resistance per unit weight of displacement = 106.3 − 105.8 × tanh(4.45 − 10.53 × Froude number + 1.66 × Prismatic coefficient + Froude number / (Longitudinal position of the center of buoyancy − 3.5)). — Real output. The model Symbolic Regression found for the resistance of a yacht hull: three inputs, one readable formula.

Run Symbolic Regression on your own data.

50,000 free credits to start: about twelve analyses at Standard depth. No credit card.

Runs from about an hour. We email you when it is done.

Start free

Diafunc's Symbolic Regression technology (research, papers, benchmarks) is at ufinq.com.

Visit ufinq.com