NCPI (National Payments Corporation of India) serves as the central agency connecting banks to facilitate interbank transfers, akin to Thailand’s NITMX. The NVIDIA GTC event reported on NCPI’s more intricate process of detecting horse account fraud that requires high processing power to efficiently identify such accounts.
The NCPI process assesses risks of individual accounts beforehand and when there is a significant amount of interbank transfers, it transfers risks akin to Google’s PageRank, which assesses the credibility of websites when they are linked. Continuous calculations assign scores to each account, indicating the level of horse account risks. Additionally, it requires identifying patterns with high risks such as multiple accounts transferring to a single account, single account dispersing funds, or self-transfers within a group.
With up to 250 million accounts to process (out of a total of 600 million accounts), linking connections becomes a massive 250m x 250m matrix. The utilization of the DGX H100 server for processing, using the cuPyNumeric library, reduced computation time from 6 hours to 30 minutes, expediting the detection of horse accounts.
Source: NVIDIA GTC Event
TLDR: NCPI utilizes high processing power to identify horse accounts efficiently, with risk assessment processes and matrix calculations, reducing computation time significantly.
Leave a Comment