🧩 3D Differentiable Logic SAT Solver: Product T-Norm Relaxation, Exact Semantic Loss and Gradient Descent
A Blender-built lab where a random planted 3-SAT formula is solved by gradient descent: a steel ball rolls on the loss surface over two logits, a rack of lamps shows satisfied clauses, and pillars show each variable probability; the product t-norm loss is compared with the exact semantic loss (weighted model count) and verified by finite differences and model counting.
A Blender-built lab where a random planted 3-SAT formula is solved by gradient descent: a steel ball rolls on the loss surface over two logits, a rack of lamps shows satisfied clauses, and pillars show each variable probability; the product t-norm loss is compared with the exact semantic loss (weighted model count) and verified by finite differences and model counting.
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install