MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641110460 A) filed by V Santhosh Kumar; and Dhiraj Sunehra on September 15, 2026, for Privacy-Preserving Hardware-Validated Split Neural Network System For Cyber-Attack Detection In Resource-Constrained Iot Devices.
Inventors include V Santhosh Kumar; and Dhiraj Sunehra.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: A privacy-preserving split neural network system for cyber-attack detection in resource-constrained IoT devices is disclosed. The system comprises a physical IoT client device executing a client-side portion of a neural network and a server executing a server-side portion thereof. The client processes locally available traffic- derived feature data to generate an intermediate activation tensor, applies controlled perturbation to the activation tensor using a differential-privacy mechanism, and transmits the perturbed activation tensor to the server without transmitting raw traffic data. The client-side neural network is transformed into multiple deployment variants comprising an uncompressed variant, a pruned variant, a reduced-precision floating-point variant, and an integer-quantized variant. The variants are converted into device-executable representations and physically evaluated on target IoT hardware using model footprint, inference latency, activation payload, communication latency, and detection performance. Hardware measurements are used to determine a deployment-suitable variant for the target device.
Disclaimer: Curated by HT Syndication.
