MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641095957 A) filed by Arukala Radhika; and Dr. M. Janga Reddy on August 07, 2026, for System And Method For Neuro-Symbolic Training Using Differentiable Consistency Constraints.
Inventors include Pondugula Kiran Kumar; Dr. Ch. Ravi; Puli Vijay Kumar; Ganga Sanuvala; Shravan Kumar B.; and Lavanya Gundu.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: A system (100) and method for neuro-symbolic training using differentiable consistency constraints are disclosed. The system (100) comprises a memory (102), a processor (104), and a communication module (106) configured to receive input training data, symbolic reasoning information, contextual dependency information, and model feedback information associated with neural network training environments. A symbolic integration module (108) transforms symbolic reasoning information into differentiable symbolic representations and injects the differentiable symbolic representations into intermediate neural network layers. A differentiable consistency enforcement module (110) generates continuously differentiable consistency constraints and propagates symbolic consistency gradients through end-to-end backpropagation. A dynamic constraint adaptation module (112) adaptively modifies constraint enforcement based on semantic divergence and optimization conditions. An adaptive neural-symbolic optimization module (114) jointly optimizes neural prediction objectives and symbolic consistency objectives to improve logical coherence, semantic consistency, interpretability, adaptive reasoning capability, and neural training stability. Fig. 1
Disclaimer: Curated by HT Syndication.
