QUALS: Corpus Equilibrium for Universal Forecasting via Pattern Quantization and Learnability Synchronization
QUALS improves zero-shot time-series foundation model training via pattern quantization and learnability synchronization, achieving better accuracy with a fraction of the data.
QUALS is a large-scale corpus equilibrium framework for time-series foundation models targeting zero-shot forecasting in domains like transportation and power grids. It combines a pattern quantization mechanism using vector quantization and uniform binning with a learnability synchronization framework that calibrates sampling weights between simple and complex motifs. Benchmarks show pre-training on QUALS delivers superior zero-shot performance even with substantially reduced training budgets.
- Uses vector quantization and uniform binning to decode heterogeneous patterns
- Calibrates sampling weights to bridge optimization gaps across motifs
- Achieves superior zero-shot forecasting with a small fraction of training data
Full article179 words · extracted from arxiv.org · click to collapse
Ubiquitous time series data across diverse domains enables critical applications in areas such as transportation systems and power grids. Recently, training foundation models on massive datasets to achieve accurate zero-shot forecasting has emerged as a major research focus. However, current studies predominantly prioritize architectural innovations while insufficiently addressing data diversity, often relying on simple data sampling strategies that fail to manage complex data distributions effectively, leading to inefficient use of training data and suboptimal performance. To address this, we propose QUALS, a large-scale time series corpus equilibrium framework. QUALS significantly enhances data efficiency, i.e., enabling existing models to achieve superior performance using only a small fraction of the original training data. Specifically, QUALS operates through two core mechanisms. First, a pattern quantization framework systematically decodes heterogeneous patterns from mixed corpora via vector quantization and uniform binning. Second, a learnability synchronization framework calibrates sampling weights for heterogeneous patterns, bridging the optimization gap between simple and complex motifs to maximize overall training efficiency. Extensive benchmarks demonstrate that pre-training on QUALS consistently achieves superior zero-shot performance, even under substantially reduced training budgets.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.20156