Published November 1, 2024
| Version v1
Preprint
Open
Information-Geometric and Topological Features of AI Behavioral Signatures: A 14-Module Mathematical Framework
Description
Description of a 14-module mathematical pipeline spanning information theory, algebraic topology, dynamical systems, differential geometry, gauge theory, and category theory. Produces 272-dimensional feature vectors expanded to 1,543 dimensions. Achieves 0.958 AUC (GPT-4) on LLM behavioral drift detection. The Hodge/sheaf decomposition module produces the single most important feature.
Notes
Part of the ABIS (Adaptive Behavioral Intelligence System) research series by CIJ Labs. Patent: GB2521226.7
Files
Files
(16.1 kB)
| Name | Size | Download all |
|---|---|---|
|
md5:b17aa1be5705cc1a239e64f6d83dc6f9
|
16.1 kB | Download |
Additional details
Related works
- Is part of
- 10.5281/zenodo.14865346 (DOI)