AI research
A research paper titled "OntoAR--HAR: Ontology--Guided Agentic RAG with Personalized Memory for Human Activity Recognition" (arXiv:2610.15292), published on October 11, 2026, introduces an ontology--guided agentic retrieval--augmented generation framework designed to resolve semantic ambiguity and hierarchical label inclusion in multi--step LLM reasoning. Traditional RAG and LLM systems typically search flat, single--level decision spaces using static population priors, leading to confusion among semantically overlapping classes. OntoAR--HAR addresses this limitation by deploying a tree--structured domain ontology that decomposes complex categorization into structured, localized decisions. The architecture integrates a static store offering node--constrained exemplars and executable signal features with an active personalized memory module. During inference, the LLM agent navigates the ontological hierarchy while dynamically synthesizing retrieved domain constraints and personalized user observations, outperforming baseline flat retrieval methods across multiple LLM backbones.