The world's fastest and most efficient AI for combinatorial optimization.
This page is a placeholder for people trying to figure out how Foxrane's optimizer actually works, and for OR-experts that are rightfully skeptical of hyperbolic AI claims. Quite soon, hopefully, we will be publishing benchmarks and technical documentation describing all our work in detail. In the meantime, the blurbs below will have to do.
Have you ever wondered whether the permutation and isometry invariance of the discrete Laplacian operator has some sort of fundamental connection to combinatorics? If so... then we think you're on to something!
Foxrane 9TAIL is a geometric deep reinforcement learning architecture that produces near-optimal plans in linear time on large combinatorial problems. Notable design elements include Universal Differential Equation networks that model input stochasticity, Laplacian eigenvector positional encodings that capture large scale structures, latent cross-attention transformer layers with subquadratic memory complexity, and a value network that can simulate long running Markov Decision Processes.
More details coming soon.