ICML 2026 · Neuro-symbolic scientific ML

SIGS Ansatz Builder

Compose an analytical candidate from typed grammar atoms, then test it against a differential equation and its prescribed conditions—entirely in your browser.

Ansatz→atoms→candidate→residuals

01 · Structure

Choose the differential equation

Equation
Problem-specific Ansatz
Conditions
PDE parameters

02 · Grammar atoms

Fill the typed slots

What is model-backed here?

The typed grammar, assembly maps, and public checkpoint are SIGS artifacts. This static release uses a deterministic browser grammar sampler so every interaction is free and reproducible. It does not label those samples as Grammar-VAE outputs. The released checkpoint remains available from the model link below.

03 · Verification

Assemble and score

Assembled candidate Load an example to begin.
Ready.
Candidate profiles
Absolute PDE residual over (x, t)

Residuals use centered finite differences on a compact public grid. They are an explanatory diagnostic, not the paper-scale Stage I search or Stage II coefficient-refinement budget.