MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202541022925 A) filed by Mastercard International Incorporated on March 13, 2025, for Methods And Systems For Training A Graph Neural Network Model Using Curriculum Learning.
Inventors include Raunak Pandey; Krisha Ketan Shah; Siddhartha Asthana; Sonia Gupta; and Ushmita Pareek.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: Methods and systems for training a GNN model using curriculum learning are disclosed. The method performed by the server system includes accessing a graph from a database. Herein, the graph includes nodes and edges. The method further includes computing a difficulty metric for each node based on a rank metric, a label metric, a feature metric, and a set of predefined weights computed for each node. Furthermore, the method includes segregating each node into difficulty classes. Each difficulty class is associated with a difficulty metric range. Herein each node is assigned a particular difficulty class if the corresponding difficulty metric of each node is within a corresponding difficulty metric range of the particular difficulty class. Moreover, the method includes assigning each difficulty class to a particular training epoch from training epochs. The method further includes training the GNN model for the training epochs using nodes assigned to each difficulty class.
Disclaimer: Curated by HT Syndication.
