Agentic Forecasting using Sequential Bayesian Updating of Linguistic Beliefs
基于序列贝叶斯更新的语义信念的代理预测
机构 * Department of Computer Science(计算机科学系) ; University of British Columbia(不列颠哥伦比亚大学)
AI总结 BLF系统通过语义信念状态、层级多试聚合和层级校准方法,在预测基准上超越了现有方法,同时具备低泄露率的回测框架。
Comments v4 adds 2 new appendices: app J evaluates a "generative" Bayesian alternative to BLF (which does worse than the "discriminative" approached used by BLF), and app K sketches how to do value of information computation to decide when to stop searching (although this idea has not been tried). v4 also adds a few more recent references
Journal ref v3 was published in ICML AI Forecasting workshop 2026 (https://forecasting-workshop.github.io/)