Intellectual Property India Publishes Patent Application for A Method To Identify At Least One Under-Represented Class In A Dataset And A System Thereof.

MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202541008064 A) filed by Bosch Global Software Technologies Private Limited; and Robert Bosch Gmbh on January 31, 2025, for A Method To Identify At Least One Under-Represented Class In A Dataset And A System Thereof..

Inventors include Abhijeet Kumar; Singh Sonam; Amit Arvind Kale; and Guruprasad M Hegde.

The application for the patent was published on August 07, 2026, under issue no. 32/2026.

Abstract: A method to identify at least one under-represented class in a dataset and a system thereof Abstract Disclosed are techniques to identify at least one under-represented class in a dataset and a system thereof. Once found, such under-represented classes will be used to identify hard samples, that is, samples of classes for which a current foundation model is failing to perform. The present disclosure uses a model agnostic approach and infuses techniques such as diffuseness score calculation and a gradcam approach. The system comprises a pre-processing module (1), a nearest neighbor module and an XAI module (5). The XAI module (5) obtains derives a set of heat maps (9) and obtains plurality of diffuseness scores for each heat map and rank a retrieved set nearest neighbor images (7) based on the diffuseness score.

Disclaimer: Curated by HT Syndication.