Why Do Time Series Models Need Long Context Windows?
Researchers have investigated the necessity of long context windows for time series forecasting models. They propose that beyond capturing long-range dependencies, these windows aid in identifying the specific data-generating process. The study demonstrates that larger context windows reduce uncertainty about the underlying process, which is crucial for accurate predictions. The findings suggest that decoupling process identification from conditional forecasting can enhance scalability without sacrificing accuracy. AI
IMPACT Provides theoretical grounding for the design of more effective time series forecasting architectures.