MUMBAI, India, Feb. 27 -- Intellectual Property India has published a patent application (202641018697 A) filed by Kalyan Revella; and Sosin Thayyaba Revella, Hyderabad, Telangana, on Feb. 19, for 'a hardware-integrated mixture-of-experts (moe) system and method for state-persistent and resource-aware neural routing.'
Inventor(s) include Kalyan Revella.
The application for the patent was published on Feb. 27, under issue no. 09/2026.
According to the abstract released by the Intellectual Property India: "The present invention discloses a system and method for modulating the internal routing logic of a Mixture-of-Experts (MoE) Transformer architecture. The system comprises a non-transitory memory storing a persistent latent-state vector (S) representing a user's cognitive or skill profile, and an infrastructure telemetry unit for generating a hardware-constrained mask (M) based on real-time physical device constraints such as CPU temperature and battery levels. A Biased Gating Module utilizes these inputs to dynamically reconfigure neural pathways, ensuring that the active neural experts are computationally viable for the host hardware while maintaining pedagogical or task-specific continuity. The invention achieves a technical effect of improved power management, prevention of hardware failure due to thermal stress, and deterministic latency on edge devices."
Disclaimer: Curated by HT Syndication.
