MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202611070235 A) filed by Ims Engineering College, Ghaziabad on June 04, 2026, for A System For Autonomous Data Labeling Using Self-Supervised Learning.
Inventors include Dr. Amit Sharma; Dr. Prabhat Kumar Srivastava; Mr. Vivek Jain; and Mr. Sanjeev Sanyal.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: The present invention discloses a system for autonomous data labeling using self-supervised learning. The system receives unlabeled, weakly labeled, or partially labeled data samples and processes them through a data ingestion module, preprocessing module, self-supervised learning module, representation embedding module, pseudo-label generation module, confidence evaluation module, label refinement engine, feedback learning module, and label repository. The self-supervised learning module learns semantic representations from data without requiring manual labels for every sample. The pseudo-label generation module assigns provisional labels using embeddings, clustering, prototype matching, similarity analysis, and model predictions. The confidence evaluation module determines reliability of generated labels and routes uncertain samples for human validation. The label refinement engine detects inconsistent, noisy, duplicate, or conflicting labels, while the feedback learning module updates the system based on verified corrections. The invention reduces manual annotation cost, improves labeling scalability, enhances dataset quality, and maintains label provenance for auditable machine learning workflows. Accompanied Drawing [FIGS. 1-2]
Disclaimer: Curated by HT Syndication.
