MUMBAI, India, July 13 -- Intellectual Property India has published a patent application (202621068925 A) filed by Symbiosis International Deemed University on June 01, 2026, for Privacy-Preserving Heterogeneous Federated Learning Framework For Depression Detection Using Logit-Level Distillation And Explainable Ai.
Inventor includes Dr. Sagar Dhanraj Pande.
The application for the patent was published on July 10, 2026, under issue no. 28/2026.
Abstract: ABSTRACT PRIVACY-PRESERVING HETEROGENEOUS FEDERATED LEARNING FRAMEWORK FOR DEPRESSION DETECTION USING LOGIT-LEVEL DISTILLATION AND EXPLAINABLE AI The present invention discloses a privacy- preserving heterogeneous federated learning framework (100) for text-based depression detection. The framework comprises four heterogeneous client nodes (110, 120, 130, 140) each training a distinct model including Logistic Regression (112), Support Vector Machine (122), Multilayer Perceptron (132), and Transformer-based BERT (142) on respective local datasets containing BERT embeddings (114). A logit-level knowledge distillation module (150) aggregates logit outputs from all client models on a shared public dataset (160) to produce averaged soft targets for training a global model (170) implemented as a fully connected neural network, without exchanging raw data or model weights. An explainable AI module (180) using SHAP provides word-level feature attribution scores for depression predictions. A cyclical refinement mechanism (190) iteratively improves the global model using explainability insights. The framework achieves 98.9 percent classification accuracy across three depression categories while preserving data privacy and ensuring prediction transparency. [
Disclaimer: Curated by HT Syndication.
