MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641096885 A) filed by Aditya Institute Of Technology And Management; Naresh Tangudu; Korla Swaroopa; and Lakshmana Rao Padala on August 11, 2026, for System And Method For Privacy-Preserving Distributed Machine Learning Optimization.
Inventors include Naresh Tangudu; Korla Swaroopa; and Lakshmana Rao Padala.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: A distributed machine learning optimization system for privacy-preserving adaptive model updating includes an edge client subsystem (100) and a cloud server subsystem (200) coupled through a network. The edge client subsystem (100) receives an inference data stream, generates a live distribution profile using an in-memory statistical sketching module (110), compares the live distribution profile with a baseline distribution profile through a drift detection module (120), determines a feature drift condition, generates a privacy-preserved statistical fingerprint through a local differential privacy module (130), and transmits the fingerprint via a communication module (140). The cloud server subsystem (200) includes an ingestion and clustering module (210), a synthetic dataset generation module (220), a model adaptation module (230), and an update deployment module (240). The statistical fingerprint excludes raw user data and model gradients, and the optimized model update is based on a synthetic proxy dataset representing feature drift.
Disclaimer: Curated by HT Syndication.
