MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641075419 A) filed by Vellore Institute Of Technology on June 18, 2026, for Carbon-Aware And Privacy-Preserving Deep Neural Network Partitioning System.
Inventors include Syamasudha Veeragandham; and Saanvi Goel.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: ABSTRACT CARBON-AWARE AND PRIVACY-PRESERVING DEEP NEURAL NETWORK PARTITIONING SYSTEM A method for partitioning a deep neural network (DNN) (200) between an edge device (100) and cloud computing infrastructure (300) is provided. The method comprises collecting device telemetry data including a device temperature value from edge sensors (104), and querying a carbon API (106) to obtain real-time carbon intensity values for cloud regions. For each candidate split layer, the method comprises computing a latency cost, a carbon emission cost, a privacy penalty, and a thermal penalty. The method comprises excluding candidate split layers within a privacy risk zone, evaluating a multi-objective cost function combining the costs and penalties, and selecting an optimal split layer and optimal cloud region minimizing the cost function. The method comprises executing layers on the edge device (100) to generate an intermediate activation tensor, transmitting the tensor to the optimal cloud region, and executing remaining layers on the cloud computing infrastructure.
Disclaimer: Curated by HT Syndication.
