Test-Time Efficient Pretrained Model Portfolios for Time Series Forecasting
This addresses computational efficiency for time series forecasting practitioners, though it appears incremental as it builds on existing ensembling and model selection techniques.
The paper tackles the problem of parameter inefficiency in time series foundation models by proposing portfolios of smaller pretrained models instead of single large models, achieving competitive performance on large-scale benchmarks with significantly fewer parameters.
Is bigger always better for time series foundation models? With the question in mind, we explore an alternative to training a single, large monolithic model: building a portfolio of smaller, pretrained forecasting models. By applying ensembling or model selection over these portfolios, we achieve competitive performance on large-scale benchmarks using much fewer parameters. We explore strategies for designing such portfolios and find that collections of specialist models consistently outperform portfolios of independently trained generalists. Remarkably, we demonstrate that post-training a base model is a compute-effective approach for creating sufficiently diverse specialists, and provide evidences that ensembling and model selection are more compute-efficient than test-time fine-tuning.