Advances in Time Series Forecasting

The field of time series forecasting is rapidly evolving, with a growing focus on developing more accurate and efficient models. Recent research has highlighted the limitations of traditional models, including their tendency to perceive non-existent patterns and their vulnerability to concept drift and temporal shifts. In response, researchers are exploring new approaches, such as the use of large language models, covariate-aware adaptation, and selective representation spaces. These innovative methods have shown promising results, including improved forecasting accuracy and reduced parameter requirements. Notably, papers such as SVTime and CoRA have introduced novel frameworks for time series forecasting, leveraging the strengths of large vision models and foundation models to achieve state-of-the-art performance. Meanwhile, papers like Lifting Manifolds to Mitigate Pseudo-Alignment in LLM4TS and Toward Reasoning-Centric Time-Series Analysis are pushing the boundaries of time series analysis, emphasizing the importance of causal structure, explainability, and human-aligned understanding. Overall, the field is moving towards more robust, adaptable, and interpretable models that can effectively handle the complexities of real-world time series data.

Sources

The Idola Tribus of AI: Large Language Models tend to perceive order where none exists

Why Do Transformers Fail to Forecast Time Series In-Context?

SVTime: Small Time Series Forecasting Models Informed by "Physics" of Large Vision Model Forecasters

Gradient-based Model Shortcut Detection for Time Series Classification

Prepared for the Unknown: Adapting AIOps Capacity Forecasting Models to Data Changes

LightSAE: Parameter-Efficient and Heterogeneity-Aware Embedding for IoT Multivariate Time Series Forecasting

Actor-Enriched Time Series Forecasting of Process Performance

CoRA: Covariate-Aware Adaptation of Time Series Foundation Models

Lifting Manifolds to Mitigate Pseudo-Alignment in LLM4TS

Toward Reasoning-Centric Time-Series Analysis

Time Series Foundation Models: Benchmarking Challenges and Requirements

Enhancing Time Series Forecasting through Selective Representation Spaces: A Patch Perspective

Tackling Time-Series Forecasting Generalization via Mitigating Concept Drift

Built with on top of