Privacy-Preserving Federated Learning over Vertically and Horizontally Partitioned Data for Financial Anomaly Detection

The project addresses a high-value, high-complexity niche: Vertical Federated Learning (VFL) for financial institutions. While horizontal FL is common, VFL (where different entities have different features for the same individuals) is significantly harder due to the requirement for Private Set Intersection (PSI) and complex alignment protocols. The quantitative signals (0 stars but 15 forks over 2.5 years) strongly indicate an academic reference implementation or a 'code-drop' associated with a research paper (arXiv:2310.19304). The 15 forks suggest that despite no public 'likes', the codebase is being utilized or audited by other researchers or institutional labs. Its defensibility is limited because it lacks a managed ecosystem or production-grade hardening, making it easily reproducible by specialized vendors like FedML, Inpher, or specialized teams within AWS/Google Cloud. However, the 'Frontier Risk' is low because general-purpose labs (OpenAI/Anthropic) are focused on LLMs, not the specialized, highly regulated, and siloed data orchestration required for inter-bank fraud detection. The project is a valuable architectural template but faces displacement by established enterprise FL platforms like FATE (Webank) or NVIDIA Flare as they move deeper into the financial vertical.