MUMBAI, India, July 30 -- Intellectual Property India has published a patent application (202641087908 A) filed by Vellore Institute Of Technology on July 17, 2026, for Continuous-Time Neural Processor For Building Energy Prediction And Hvac Control.
Inventors include Muchenedi Hari Kishor; Aryajeet Jha; and Garvit Joshi.
The application for the patent was published on July 24, 2026, under issue no. 30/2026.
Abstract: ABSTRACT CONTINUOUS-TIME NEURAL PROCESSOR FOR BUILDING ENERGY PREDICTION AND HVAC CONTROL The present disclosure provides a training method (300) for a continuous-time neural processor (108) deployed in a building IoT energy management system (100). The training method (300) comprises independently initializing a learnable temporal parameter for each processing unit from a log-normal distribution with values constrained to a numerically stable interval during parameter initialization (302), and training the continuous-time neural processor (108) using a composite loss formation (312) comprising a mean squared error term on building energy prediction and a temporal diversity enforcement term of the form negative lambda multiplied by variance of all per-unit temporal parameters across all processing layers, wherein lambda is a positive diversity enforcement coefficient. The training method (300) produces, as a deterministic and statistically reproducible outcome across multiple independent training runs, a three-tier temporal specialization wherein fast units (322), medium units (324), and slow units (326) emerge without explicit architectural specification. (Fig. 1)
Disclaimer: Curated by HT Syndication.
