MUMBAI, India, May 29 -- Intellectual Property India has published a patent application (202641063427 A) filed by M. Kumarasamy College Of Engineering, Karur, Tamil Nadu, on May 19, for 'a method for providing multi scale explainability of deep neural network model predictions.'
Inventor(s) include Mrs. S. Ananthi; and T. Naveenraj.
The application for the patent was published on May 29, under issue no. 22/2026.
According to the abstract released by the Intellectual Property India: "The present invention discloses a multi-scale explainability framework that systematically reveals how deep neural networks make decisions at every layer. The core innovation lies in training interpretable surrogate models for each hidden layer, approximating the relationship between layer activations and the final prediction. These surrogates are organized into a hierarchical explanation tree, allowing users to interactively explore and understand the decision process layer by layer. Fidelity metrics are computed to ensure that the surrogates accurately reflect the base model's internal logic. The system integrates a web-based dashboard providing visualizations such as decision rules, feature importance graphs, and counterfactual scenarios. LayerLens is designed for both PyTorch and TensorFlow models, enabling its application in real-time environments, enhancing debugging, safety audits, and regulatory compliance of machine learning models."
Disclaimer: Curated by HT Syndication.
