MUMBAI, India, Oct. 12 -- Intellectual Property India has published a patent application (202641118397 A) filed by Rns Institute Of Technology, Bengaluru; Ms. Yashasvi B. Nassistant Professor; Mr. Chetan Ghatage Assistant Professor; Rakesh Ms Rncs; Sankalp Raj Rncs; Rohith O Rncs; and Navaneeth K Rncs on October 03, 2026, for Ai-Based System For Detecting Data Leakage In Image Classification Datasets Using Deep Learning Techniques.
Inventors include Ms. Yashasvi B. Nassistant Professor; Mr. Chetan Ghatage Assistant Professor; Rakesh Ms Rncs; Navaneeth K Rncs; Sankalp Raj Rncs; and Rohith O Rncs.
The application for the patent was published on October 09, 2026, under issue no. 41/2026.
Abstract: With the rapid growth of machine learning applications, dataset quality has become a critical factor affecting model performance. One major issue is data leakage, where similar or identical data appears across training and testing datasets, leading to misleading accuracy and unreliable models. The proposed system introduces an AI-based automated data leakage detection framework that uses deep learning models such as ResNet, CLIP, and Vision Transformers to extract feature embeddings from images. These embeddings are analyzed using similarity measures to detect duplicate, near-duplicate, and semantically similar images. The system performs multi-level leakage detection, including image-level, label-level, and metadata-level analysis, and supports multiple datasets such as CIFAR-10, COCO, and medical datasets. Efficient similarity search techniques are used to ensure scalability. The proposed system enhances dataset reliability, improves model generalization, and reduces risks associated with biased or leaked data in machine learning pipelines.
Disclaimer: Curated by HT Syndication.
