MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641101749 A) filed by Madhankumar C; Ms Sivasakthi C - Nehru Arts And Science College; Mr M. Dinesh - Erode Sengunthar Engineering College; Mr Raya Pawan Kumar - Giet University; Ms M Revathi - Rathinam Technical Campus; Dr. Prathima Mabel J - Dayananda Sagar College Of; and Dr. Renu Puri - Chandigarh Engineering College on August 23, 2026, for A Deep Learning Framework For Real-Time Predictive Analytics And Explainable Decision Intelligence Using Multi-Modal Foundation Models.
Inventors include Ms Sivasakthi C - Nehru Arts And Science College; Mr M. Dinesh - Erode Sengunthar Engineering College; Mr Raya Pawan Kumar - Giet University; Ms M Revathi - Rathinam Technical Campus; Dr. Prathima Mabel J - Dayananda Sagar College Of; and Dr. Renu Puri - Chandigarh Engineering College.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: A Deep Learning Framework for Real-Time Predictive Analytics and Explainable Decision Intelligence Using Multi-Modal Foundation Models ABSTRACT The present invention relates to a Deep Learning Framework for Real-Time Predictive Analytics and Explainable Decision Intelligence Using Multi-Modal Foundation Models, designed to provide intelligent, real-time, and explainable analysis of heterogeneous data generated from multiple sources. Conventional predictive analytics systems generally process a limited type of data and often depend on separate machine learning models for individual data modalities, resulting in fragmented analysis, increased computational complexity, limited adaptability, and reduced transparency in decision- making. The proposed invention addresses these limitations through an integrated deep learning framework capable of processing and combining multi-modal data using foundation-model- based intelligence. The proposed framework comprises a multi-modal data acquisition layer, data preprocessing and synchronization modules, modality-specific encoders, a multi-modal fusion engine, a foundation model layer, a real-time predictive analytics engine, an explainable decision intelligence module, a confidence estimation module, and an interactive decision support interface. The framework is configured to receive heterogeneous information including structured numerical data, text, images, audio, video, sensor streams, time-series data, and contextual information. The received data is normalized, synchronized, transformed into suitable representations, and supplied to corresponding modality-specific processing components. The modality- specific encoders transform heterogeneous input data into meaningful feature representations. The multi-modal fusion engine combines the extracted representations into a unified contextual representation while preserving relationships between different data modalities. The resulting representation is processed by one or more multi-modal foundation models to identify complex patterns, correlations, temporal relationships, and contextual dependencies that may not be effectively identified using conventional single-modal predictive models. The real-time predictive analytics engine generates predictions, risk scores, classifications, forecasts, anomaly indicators, and other analytical outputs based on continuously updated multi-modal information. The framework is capable of supporting streaming data analysis and dynamically updating predictions when new information becomes available. This enables timely decision-making in applications where operational conditions continuously change. A key feature of the invention is the explainable decision intelligence module, which converts model outputs into understandable explanations. The module identifies influential input features, relevant data modalities, contributing patterns, confidence levels, and decision factors associated with a prediction. The generated explanations may be presented using feature importance scores, attention-based representations, visual indicators, textual explanations, confidence values, or other interpretable formats. The confidence estimation module evaluates the reliability of generated predictions by considering model confidence, data quality, modality agreement, historical performance, and uncertainty characteristics. When prediction confidence falls below a configurable threshold, the system may request additional information, initiate further analysis, flag the prediction for human review, or select an alternative analytical pathway. The framework further incorporates a continuous feedback mechanism through which actual outcomes, user decisions, expert feedback, and subsequent observations are used to improve predictive performance and decision intelligence. The system can therefore adapt to changing data distributions and operational environments without requiring complete reconstruction of the analytical architecture. The proposed invention provides a unified framework for real-time multi-modal learning, predictive analytics, explainable artificial intelligence, uncertainty-aware decision support, and continuous model improvement. The framework can be applied to smart healthcare, financial analytics, industrial monitoring, cybersecurity, intelligent transportation, education, smart cities, business intelligence, and other domains requiring rapid and transparent data-driven decision- making. The invention thereby improves predictive accuracy, contextual understanding, real-time responsiveness, interpretability, decision reliability, and adaptability while providing a scalable architecture for deploying multi-modal foundation-model intelligence in real-world applications.
Disclaimer: Curated by HT Syndication.
