Decentralized Online Multi-Agent Visual Question Answering

Ghulam Ahmed Ansari, Amrita Saha, Srikanth G. Tamilselvam

Abstract: This invention introduces a decentralized multi-agent system for Visual Question Answering (VQA) in an online learning setting. It integrates neural-symbolic reasoning, dynamically inducing executable programs over operators (e.g., find, relate) and concepts (e.g., man, next to). Unlike traditional VQA systems that rely on static datasets and centralized processing, this approach enables scalable, modular, and adaptive learning. Agents collaborate to answer questions while minimizing cascading errors, making the system robust to unseen data and real-world scenarios.