MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641096401 A) filed by Velammal Institute Of Technology; Mr. Raja R; Mr. Sundaresan B; Dr. Venkadesh R; Jaikiran J; Ajith Kumar A; Mano M; and Yaswanth Sai R on August 10, 2026, for Learning Based Predictive Autoscaling System Using Federated Learning On Cloud Infrastructure.
Inventors include Mr. Raja R; Mr. Sundaresan B; Dr. Venkadesh R; Jaikiran J; Ajith Kumar A; Mano M; and Yaswanth Sai R.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: The growing demand for scalable and efficient cloud resource management has highlighted the limitations of traditional reactive auto-scaling systems, which trigger scaling actions only after workload thresholds are breached, leading to performance degradation and resource inefficiency. To address these challenges, the present invention proposes a machine learning-based predictive autoscaling system that integrates federated learning with Long Short-Term Memory (LSTM) neural networks for proactive and privacy-preserving cloud resource management on AWS. Each EC2 instance independently trains a local LSTM model using its own CloudWatch resource utilization metrics, including CPU usage, memory consumption, and network traffic, without sharing raw data with any central server. Only trained model parameters are transmitted to a central AWS Lambda-based aggregation service, which applies the Federated Averaging (FedAvg) algorithm to produce a global LSTM model. This global model forecasts future workload demands, and based on these predictions, the system proactively triggers scaling actions through AWS Auto Scaling Groups before resource contention occurs. A continuous retraining loop ensures the system adapts to evolving workload patterns over time. The architecture integrates AWS CloudWatch for telemetry collection, AWS Lambda for federated aggregation, and AWS Auto Scaling Groups for dynamic instance management. This integration ensures scalability, data privacy, reduced latency, and optimized resource utilization. Overall, the proposed system provides a comprehensive solution for intelligent cloud resource management by eliminating reactive delays, preserving data privacy across distributed nodes, and improving application performance through predictive autoscaling. Keywords: Machine Learning, Federated Learning, Predictive Autoscaling, AWS, LSTM, CloudWatch, Auto Scaling Groups, AWS Lambda, FedAvg, Cloud Resource Management, Workload Prediction, Distributed Systems.
Disclaimer: Curated by HT Syndication.
