MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202641048346 A) filed by L Jabasheela; Dr. P. Deepa; Dr. V. Sathiya; Mr. Gunasekar S D; and Mr. R. Kishant on April 16, 2026, for Mojonet: An Experimental Deep Learning Framework.
Inventors include Dr. P. Deepa; Dr. V. Sathiya; Mr. Gunasekar S D; Mr. R. Kishant; L Jabasheela; Dr. R. Salini; Ravindran U; Lincy Jemina S; Sophana Jennifer S; and Veera Manikandan K.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: Mojonet is an experimental deep learning framework written entirely in the Mojo programming language. In most existing frameworks, the user-facing code is in Python and the performance code is in C++/CUDA, creating a language boundary that limits whole-program optimisation. Mojonet removes this boundary by placing all components tensor operations, kernels, automatic differentiation, layers, optimisers, and training utilities in one language and one compiler pipeline (MLIR). The underlying algorithms are standard techniques from the literature and are not claimed as new. The contribution is the single-language system architecture and the empirical demonstration of its performance: 94% of PyTorch throughput on typical workloads, with 8–12% less peak memory.
Disclaimer: Curated by HT Syndication.
