MUMBAI, India, Aug. 1 -- Intellectual Property India has published a patent application (202441005993 A) filed by Samsung Electronics Co. Ltd., Gyeonggi, Republic of Korea, on Jan. 30, 2024, for 'method and system for fast learning neural architecture search (nas) with hardware-informed surrogates.'

Inventor(s) include Miriyala Srinivas Soumitri; Suhas P K; Utsav Tiwari; Vajrala Sowmya Lahari; and Vikram Nelvoy Rajendiran.

The application for the patent was published on Aug. 1, under issue no. 31/2025.

According to the abstract released by the Intellectual Property India: "Embodiments of the present disclosure disclose a method (600) and system (111) for optimization of an artificial intelligence (AI) model for deployment which includes implementing a pre-trained AI model (105) for performing at least one task to evaluate one or more on-device key performance indicators (KPIs) and identifying (604) at least one sub-optimal block in the AI model (105), based on one or more on-device KPIs. Method (600) includes retrieving (606) plurality of alternative blocks corresponding to each sub-optimal block and training (608) plurality of alternative blocks based on input and output features of corresponding sub-optimal block to generate plurality of surrogate blocks for each sub-optimal block. Further, it includes applying (610) at least one search strategy to select the best combination of surrogate blocks from a plurality of surrogate blocks for replacing corresponding plurality of the sub-optimal blocks and replacing (612) each sub-optimal block with the respective selected surrogate block to generate an optimized AI model (115)."

Disclaimer: Curated by HT Syndication.