MUMBAI, India, May 1 -- Intellectual Property India has published a patent application (202641051630 A) filed by Vellore Institute Of Technology, Vellore, Tamil Nadu, on April 23, for 'hybrid knowledge distillation and quantization framework for lightweight defect classification.'
Inventor(s) include Arumuga Arun R; G. Manikandan; Ponnarasan V; and Sarathy B.
The application for the patent was published on May 1, under issue no. 18/2026.
According to the abstract released by the Intellectual Property India: "The present invention relates to a computer-implemented system (200) and method for generating a lightweight deep learning model using a hybrid knowledge distillation and quantization framework. The system (200) comprises a preprocessing module configured to standardize input images, a teacher model module configured to generate predictive probability distributions, and a static probability generation module configured to produce temperature-scaled softened outputs. A student model is trained using a knowledge distillation module employing a composite loss function comprising hard-label and soft-label components derived from precomputed probabilities. A serialization module is configured to extract the trained model while excluding optimizer state variables. A quantization module converts the trained model into reduced precision representations. The system (200) enables coordinated integration of static probability-based distillation, optimizer-decoupled serialization, and post-training quantization within a unified processing framework."
Disclaimer: Curated by HT Syndication.
