The paper presents a new approach to crowdsourcing, leveraging self-supervised learning and a novel aggregation scheme. Instead of averaging estimates from crowdworkers, it adapts weights based on their previous estimates. This method is more accurate when skills vary across crowdworkers or their estimates correlate. The paper also discusses the limitations of existing algorithms like expectation maximization, especially when complex models are needed. The proposed approach can handle such complexities, making it more effective for practical crowdsourcing systems.
Publication date: 24 Jan 2024
Project Page: https://arxiv.org/abs/2401.13239
Paper: https://arxiv.org/pdf/2401.13239
