Bridge-TS: Improving Time Series Imputation with Enhanced Priors

This paper addresses the challenge of time series imputation, a crucial task in various domains. It innovates by focusing on the prior knowledge used in generative models. The core contribution lies in the design of 'expert prior' and 'compositional priors' to guide the generation process, leading to improved imputation accuracy. The use of pre-trained transformer models and the data-to-data generation approach are key strengths.