MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202621084404 A) filed by National Payments Corporation Of India on July 09, 2026, for Multi-Model Fusion And Streamed Matrix Multiplication For Transaction Network Analysis.
Inventors include Prashant Devadiga; Nitin Kukreja; Chandra Bhushan; Harsh Sharma; and Suvradip Paul.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: ABSTRACT MULTI-MODEL FUSION AND STREAMED MATRIX MULTIPLICATION FOR TRANSACTION NETWORK ANALYSIS A system for real-time anomaly detection in payment transaction networks is described. In one example, the system comprises a streaming ingestion layer, a GPU-resident graph-state management layer, a subgraph pruning layer, an anomaly-detection computation layer, and a scoring and flagging layer. Transaction deltas are applied in-place to a compressed-sparse-row adjacency matrix in GPU high-bandwidth memory without host-to-device transfer between micro-batch windows. A finite k-hop bidirectional sparse matrix multiplication engine on GPU cores computes forward and backward hop matrices with per-hop binarisation. A Focused Colored Graph Model restricts computation to a 5-hop neighbourhood of known suspicious accounts. Candidate accounts are scored in parallel GPU chunks and flagged when a risk score exceeds a predefined threshold. Single-node and multi-node GPU deployments are supported via a workload scheduler. FIG. 1.
Disclaimer: Curated by HT Syndication.
