MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202611101709 A) filed by Panipat Institute Of Engineering And Technology Piet on August 22, 2026, for System And Method For Adversarial Attack Detection In Machine Learning Pipelines Using Input Perturbation Analytics.
Inventors include Piyush; Anju Arya; and Dr. Vaishali Mehta.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: [046] The present invention discloses a system and method for adversarial attack detection in machine learning pipelines using input perturbation analytics. The invention employs an input acquisition module, a perturbation generation engine, a feature consistency analyzer, a prediction behavior analyzer, a multi-stage correlation engine, an adaptive learning module, a response orchestration module, and a security repository to continuously monitor machine learning operations for adversarial manipulation. Controlled perturbation variants are generated and compared with original inputs to evaluate feature consistency, prediction stability, confidence variation, and behavioral divergence. Multiple analytical outputs are correlated to produce a comprehensive adversarial threat score, enabling accurate detection of malicious input manipulation while minimizing false positives. Adaptive learning continuously refines detection parameters using operational feedback, thereby improving long-term robustness against evolving attack strategies. Upon detecting adversarial activity, the system automatically executes security responses including alert generation, secure inference routing, input quarantine, forensic logging, and policy enforcement. The invention provides scalable, computationally efficient, and adaptive protection for machine learning pipelines deployed across cloud, edge, enterprise, and industrial environments. Accompanied Drawing [FIGS. 1- 2]
Disclaimer: Curated by HT Syndication.
