Multiple noisy label distribution propagation

Abstract

With the growing demand for labeled data in AI, crowdsourcing has emerged as a cost-effective solution for data labeling. However, crowdsourced labels often suffer from inconsistencies due to varying quality from different labelers. To address this, Multiple Noisy Label Distribution Propagation (MNLDP) has been proposed to improve label accuracy by propagating labels among similar data points. Despite its effectiveness, MNLDP struggles with complex or noisy datasets, highlighting the need for more robust methods to enhance label integration in crowdsourced data. This report explores these limitations and suggests potential directions for overcoming them.