Intellectual Property India Publishes Patent Application for Federated Deep Learning System For Privacy-Preserving Intrusion Detection Across Distributed Heterogeneous Computer Networks And Devices

MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202621107858 A) filed by Prof. Bhagyashree Lambture; Prof. Ujjwala Nivas Salunkhe; Prof. Dr. Aparna Tiwari; Prof. Radhika Adki; Dr. Jagannath Nalavade; Prof. Mandar Deepak Joshi; Ms. Saloni Vipulkumar Thakar; and Dr. Umesh Kulkarni on September 08, 2026, for Federated Deep Learning System For Privacy-Preserving Intrusion Detection Across Distributed Heterogeneous Computer Networks And Devices.

Inventors include Prof. Bhagyashree Lambture; Prof. Ujjwala Nivas Salunkhe; Prof. Dr. Aparna Tiwari; Prof. Radhika Adki; Dr. Jagannath Nalavade; Prof. Mandar Deepak Joshi; Ms. Saloni Vipulkumar Thakar; and Dr. Umesh Kulkarni.

The application for the patent was published on October 30, 2026, under issue no. 44/2026.

Abstract: ABSTRACT The invention discloses a federated deep learning system for privacy-preserving intrusion detection across computer networks and devices. The system comprises a coordinating server and client nodes, each having a traffic acquisition interface, preprocessing module, feature extraction module, local training engine, privacy module, inference engine, communication interface, and alert generator. Each client trains a local intrusion detection model using retained network packets, flow records, event logs, or device telemetry, without transferring raw data outside its security boundary. Local parameter updates are clipped and protected using differential privacy, encryption, masking, secret sharing, secure aggregation, or combinations thereof. The coordinating server authenticates clients, selects participants, detects anomalous or poisoned updates, and combines accepted contributions through robust weighted aggregation to produce an updated global model. The global model is redistributed to authorized clients for local classification of normal behavior, known attacks, and previously unseen anomalies. Client selection and model processing adapt to heterogeneous datasets, computing resources, bandwidth, availability, and security conditions. Drift monitoring and feedback support repeated training as network behavior changes. The system generates local alerts with classification, confidence, severity, affected resource, and response information while limiting shared incident details. Accordingly, collaborative threat intelligence is achieved without centralized storage of sensitive traffic datasets.

Disclaimer: Curated by HT Syndication.