MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641124516 A) filed by Lords Institute Of Engineering And Technology Autonomous on October 17, 2026, for Method For Adaptive Trust Scoring Of Distributed Inference Nodes Using Multi-Modal Hardware Performance Counters In Heterogeneous Compute Clusters..
Inventors include Dr Mohammed Tajuddin; Dr Moksud Alam Mallik; Dr G. M. Sayeed Ahmed; Dr Khaja Mizbahuddin Quadry; Dr Abdul Rasool Md; Dr T. K. Shaik Shavali; Dr Shaik Qadeer; Dr Abdul Mateen Ahmed; Dr Ayesha Mariyam; and Dr Amairullah Khan Lodhi.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: A method and system for adaptive trust scoring of distributed inference nodes in a heterogeneous compute cluster are disclosed. Hardware performance counter telemetry spanning multiple modalities — including processor-execution, memory/cache, accelerator-utilization, and power/thermal counters — is collected from inference nodes having differing hardware architecture classes during runtime execution of inference tasks. The counter observations are normalized against hardware- class-specific baselines so that measurements from heterogeneous hardware become comparable, and modality-specific deviation values are computed against node- specific and peer-node baselines. The deviation values are fused, using confidence weights reflecting counter availability and reliability, into a fused deviation score from which an instantaneous trust observation is derived. A trust score for each node is adaptively updated using an update rate that depends on measurement confidence and on detection of a change-point condition, with asymmetric rates for trust-decreasing and trust-increasing updates, and the node is assigned to a hysteresis-based trust state. The trust state and score are used to control distributed inference scheduling, task verification, and node quarantine, enabling continuous, runtime, hardware-aware reliability assessment across heterogeneous compute clusters.
Disclaimer: Curated by HT Syndication.
