MUMBAI, India, Jan. 9 -- Intellectual Property India has published a patent application (202541134506 A) filed by Indian Institute Of Technology, Hyderabad, Telangana, on Dec. 31, 2025, for 'a method and a system for determining an optimal kernel size in a convolutional neural network (cnn) model.'
Inventor(s) include Tavva Dinesh Reddy; Chiruvella Gagan Chandra Sai Ram; Boreddy Sri Sai Abhinav Reddy; and Anjana A M.
The application for the patent was published on Jan. 9, under issue no. 02/2026.
According to the abstract released by the Intellectual Property India: "The present subject matter discloses a method for determining an optimal kernel size in a CNN model. The method includes defining a plurality of parameters of the CNN model. The method includes selecting a range of kernel sizes. The method includes creating a distinct variant for each kernel size. The method includes training the distinct variant for each kernel size on a target dataset to generate a unique set of trained weights for each distinct variant. The method includes injecting one or more random bit-flip errors in at least one trained weight across a range of BERs. The method includes evaluating a functional performance of the at least one trained weight corrupted with the one or more random bit-flip errors on a validation set to generate performance data. The method includes comparing a performance of the distinct variant of each kernel size at each BER to identify a high performing kernel size amongst the one or more kernel sizes with respect to a robustness and a stability."
Disclaimer: Curated by HT Syndication.
