MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202641073532 A) filed by Andhavarapu Bhanusri; and Dr. Mogalla Shashi on June 12, 2026, for An Ai-Driven Personalized Federated Emotion Analytics System For Real-Time Human Emotion Detection From Eeg Brainwave Patterns.
Inventors include Andhavarapu Bhanusri; and Dr. Mogalla Shashi.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: ABSTRACT The present invention discloses an AI-driven, decentralized, and personalized federated emotion analytics system designed for real-time human emotion detection from Electroencephalogram (EEG) brainwave patterns while preserving absolute user data privacy. Conventional emotion recognition architectures depend heavily on centralized cloud processing systems, which require the raw, sensitive neurophysiological data of a user to be transmitted continuously over external networks. This reliance on centralized servers introduces severe data privacy vulnerabilities, creates bandwidth bottlenecks, and yields high latencies that make real- time analytical intervention unfeasible. The present invention resolves these fundamental limitations by establishing an on-device, localized edge-computing framework that processes raw multi-channel EEG signals, extracts complex spatio-temporal neuro-features locally, and updates a localized personalized model without exposing the underlying raw data. To achieve robust generalization across heterogeneous demographic groups without compromising individual privacy, the present architectural framework implements an asynchronous federated learning protocol mediated by a secure global aggregation server. Instead of transmitting raw EEG timeseries data, individual edge nodes securely upload encrypted, localized model gradients and hyperparameter updates to the central aggregator via a differentially private cryptographic channel. The global aggregation server synthesizes these decentralized updates using a modified federated averaging coefficient that accounts for data non-independent and identically distributed (non-IID) variations across different subjects, updating a global foundational model. This updated global model structure is subsequently back-propagated to the edge devices, where a localized personalization layer adapts the global weights to fit the unique neurological baselines, cognitive profiles, and physiological variations of the specific user. The decentralized architecture ensures that the system continuously adapts to the user's changing neurological baselines over extended periods, effectively neutralizing the common issue of cross-subject variability that plagues traditional emotion recognition systems. By incorporating an on-device reinforcement feedback loop, the system dynamically refines its personalization layer based on implicit and explicit contextual responses, yielding highly precise, real-time affective classification. The practical deployment of this invention extends across critical domains, including real-time mental health tracking, neuro-adaptive learning environments, high-stress operational monitoring for aviation and military personnel, and immersive human-computer interfaces. By combining localized edge processing, personalized deep neural layers, and cryptographically secure federated learning, the invention establishes a scalable, low-latency infrastructure for continuous, privacy-preserving neurological emotion analytics.
Disclaimer: Curated by HT Syndication.
