The field of time series forecasting is witnessing significant developments with the integration of large language models (LLMs). Researchers are exploring the potential of LLMs to improve forecasting accuracy and robustness. One notable direction is the use of non-causal, bidirectional attention encoder-only transformers, which have shown state-of-the-art performance in certain tasks. Additionally, there is a growing interest in multimodal time series forecasting, where LLMs are used to incorporate visual data, such as satellite imagery, into forecasting models. Another area of research focuses on developing more efficient and scalable time series foundation models, which can be fine-tuned for specific downstream tasks. Noteworthy papers include Output Scaling: YingLong-Delayed Chain of Thought, which presents a joint forecasting framework for time series prediction, and Large Language models for Time Series Analysis: Techniques, Applications, and Challenges, which provides a systematic review of pre-trained LLM-driven time series analysis.
Advances in Time Series Forecasting with Large Language Models
Sources
Output Scaling: YingLong-Delayed Chain of Thought in a Large Pretrained Time Series Forecasting Model
Large Language models for Time Series Analysis: Techniques, Applications, and Challenges
NSW-EPNews: A News-Augmented Benchmark for Electricity Price Forecasting with LLMs
Can Time-Series Foundation Models Perform Building Energy Management Tasks?
Prioritizing Alignment Paradigms over Task-Specific Model Customization in Time-Series LLMs
FAA Framework: A Large Language Model-Based Approach for Credit Card Fraud Investigations
HAELT: A Hybrid Attentive Ensemble Learning Transformer Framework for High-Frequency Stock Price Forecasting
Extracting transient Koopman modes from short-term weather simulations with sparsity-promoting dynamic mode decomposition
Multi-Scale Finetuning for Encoder-based Time Series Foundation Models
SKOLR: Structured Koopman Operator Linear RNN for Time-Series Forecasting
PIPE: Physics-Informed Position Encoding for Alignment of Satellite Images and Time Series
ss-Mamba: Semantic-Spline Selective State-Space Model
Explain First, Trust Later: LLM-Augmented Explanations for Graph-Based Crypto Anomaly Detection
Advanced Prediction of Hypersonic Missile Trajectories with CNN-LSTM-GRU Architectures
Conditional Generative Modeling for Enhanced Credit Risk Management in Supply Chain Finance