Advances in Time Series Forecasting

The field of time series forecasting is moving towards more generalizable and adaptable models. Recent developments have focused on improving the accuracy and efficiency of forecasting systems, particularly in scenarios where data is limited or uncertain. The use of pre-trained models, attention mechanisms, and resolution-aware retrieval strategies have shown promising results in achieving state-of-the-art performance across various benchmarks. Notably, the integration of fairness and epidemic dynamics with vaccine logistics has led to superior outcomes in epidemic forecasting. Furthermore, the development of lightweight foundation models has made it possible to deploy forecasting systems in resource-constrained environments. Some noteworthy papers in this area include: Chronos-2, which delivers state-of-the-art performance across three comprehensive benchmarks. Resolution-Aware Retrieval Augmented Zero-Shot Forecasting, which significantly outperforms traditional forecasting methods in microclimate forecasting. Proactive and Fair Epidemic Resource Allocation, which prevents more than 2 million infections and 30,000 deaths in six months. A Conditional Diffusion Model, which achieves a relative Mean Absolute Error of 0.94% in battery capacity degradation forecasting. A Climate-Aware Deep Learning Framework, which reliably outperforms statistical baselines in RSV forecasting. SEMPO, which reduces pre-training data scale and model size while achieving strong generalization. QKCV Attention, which enhances forecasting accuracy of attention-based models across diverse real-world datasets.

Sources

Chronos-2: From Univariate to Universal Forecasting

Resolution-Aware Retrieval Augmented Zero-Shot Forecasting

Proactive and Fair Epidemic Resource Allocation Through an Integrated Supply Chain Framework: Insights from a COVID-19 Study

A Conditional Diffusion Model for Probabilistic Prediction of Battery Capacity Degradation

A Climate-Aware Deep Learning Framework for Generalizable Epidemic Forecasting

SEMPO: Lightweight Foundation Models for Time Series Forecasting

QKCV Attention: Enhancing Time Series Forecasting with Static Categorical Embeddings for Both Lightweight and Pre-trained Foundation Models

Built with on top of