MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641109957 A) filed by Dr. Binoy Balan K; Dr. Murugananth Gopal Raj; Mrs. Sandhya P. S.; and Mrs. S. Varunadevi on September 14, 2026, for System And Method For Generating Centrality-Parameterized N-Graph Topological Explanations For Graph-Based Artificial Intelligence Predictions.
Inventors include Dr. Binoy Balan K; Dr. Murugananth Gopal Raj; Mrs. Sandhya P. S.; and Mrs. S. Varunadevi.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: The present invention relates to a computer-implemented system and method for generating structured, local, and interpretable topological explanations for predictions produced by graph-based artificial intelligence systems. A prediction-specific graph instance is received together with associated node and/or edge attributes, and a graph-based artificial intelligence predictor generates a prediction for the instance. A selected centrality function is applied to vertices and/or edges of the graph to obtain structural importance values. A threshold parameter is then applied to construct a centrality-guided subgraph representing a candidate evidence region. A covering family of subgraphs is used to derive a lower approximation comprising covering components completely contained within the selected region and an upper approximation comprising covering components intersecting the selected region. A boundary region is obtained from the difference between the upper and lower approximations. These approximation structures are used to generate a smallest graph topology containing the explanation-generating components, thereby forming a structured explanation object comprising the prediction, selected evidence region, structurally certain core, relevant structural context, ambiguity boundary, and topological state. The invention further permits analysis of threshold stability, comparison of explanation states generated by different centrality functions, and empirical validation through controlled perturbations of explanatory and non-explanatory graph regions. The system thereby provides a formal structural explanation layer that can operate with graph neural networks and other graph-based prediction systems.
Disclaimer: Curated by HT Syndication.
