Knowledge Manifold: Riemannian geometry maps scientific literature and invents new research

Tomonaga Okabe and Kazuhiko Komatsu have proposed a novel framework called the Knowledge Manifold, which treats a corpus of scientific documents as points on a Riemannian manifold. The pipeline begins by converting each paper into a character-level n-gram TF-IDF vector (4-7 grams, up to 250,000 features, L2-normalized) and embedding them in a 2D knowledge map via constrained stress minimization. From this map, Smoothed Particle Hydrodynamics (SPH) interpolation estimates the feature vector at any arbitrary query point, enabling linguistic characterization. Directional knowledge gradients (0°, 45°, 90°) are computed using inner product and cosine similarity, and a Gaussian Process Regression (GPR) model with a Constant × RBF + White kernel on a 10-dimensional SVD projection provides Bayesian posterior means and uncertainty estimates. Finally, geodesics are found by minimizing a discrete Riemannian path energy derived from the SPH-induced metric tensor, using L-BFGS-B with seven initial-path candidates.

Applied to a corpus of 20 papers on fiber-reinforced composite materials and aerospace structural mechanics, the semantic map recovers meaningful research clusters, and geodesic paths reveal natural conceptual bridges between distant topics. More intriguingly, SPH/GPR interpolation enables the generation of 'virtual knowledge' – hypothetical paper abstracts describing unstudied but geometrically predicted research directions. This suggests a future where AI not only retrieves existing knowledge but actively proposes novel interdisciplinary connections, potentially accelerating scientific discovery.