Rethinking Time-Series Imputation as Conditional Inference along Temporal Evolution

Real-world time-series data often suffer from missing observations, hindering long-range temporal modeling. However, most existing imputation methods formulate imputation as conditional reconstruction over limited context, which restricts temporal information propagation and fails to explicitly model temporal evolution. To overcome this limitation, we propose the Conditional Temporal Inference Paradigm (CTIP), which formulates time-series imputation as conditional inference along temporal evolution. Under this paradigm, we introduce CBiT, which leverages a history compression mechanism to encode long-range history into a compact latent space for history-conditioned temporal imputation. In addition, we adopt a partitioned modeling strategy that distinguishes historical context and temporal imputation targets with only linear-time complexity. Extensive experiments on multiple public benchmarks show that CBiT improves imputation accuracy by reducing Masked MAE and Masked RMSE by 27.3% and 18.6%, respectively, across different missing rates.

Real-world time-series data, such as sensor readings, energy records, and weather measurements, often contain missing values. These gaps can make it harder to understand how a system changes over time and may reduce the reliability of later analysis or decision-making. Many existing methods fill in missing values mainly by using nearby observed data, which can overlook broader historical patterns. This paper proposes the Conditional Temporal Inference Paradigm, or CTIP, a new way to think about time-series imputation. Instead of treating missing values as isolated gaps to be locally repaired, CTIP views them as values that should be inferred from the temporal evolution of the data. Based on this idea, we introduce a model named CBiT, which uses historical information to guide the recovery of missing values in the current incomplete segment. Experiments on several public time-series datasets show that this approach can produce more accurate and stable imputations than existing methods, while keeping the computational cost low. This makes the method useful for applications where reliable time-series data is important, such as monitoring systems, energy management, and environmental analysis.