ICLR 2026 Novelty Validation Report

Abstract

Assessing the novelty of research ideas is critical yet labor-intensive due to the vast volume of prior work. We propose a novel automated agent designed to streamline this process through idea retrieval, semantic retrieval, and evidence verification. We deployed this agent to analyze a dataset comprising the top 200 papers from the ICLR 2026 OpenReview. Our findings and the system's performance are demonstrated as follows.