The field of time series forecasting and analysis is rapidly advancing with the development of new models and techniques. Recent research has focused on improving the accuracy and efficiency of time series forecasting, particularly in areas such as network traffic forecasting, multivariate time series forecasting, and anomaly detection. One notable trend is the use of foundation models, which have demonstrated strong zero-shot and few-shot generalization capabilities across diverse domains. Additionally, diffusion-based models have shown promise in generating multi-scale mobile network traffic and improving time series forecasting performance. Other innovative approaches include the use of wavelet decomposition, attention mechanisms, and contrastive forecasting frameworks. Noteworthy papers in this area include Time-Series Foundation Models for ISP Traffic Forecasting, which achieves state-of-the-art performance in network traffic forecasting, and SimDiff, which proposes a simpler yet better diffusion model for time series point forecasting. Overall, these advances have the potential to significantly impact various applications, from network monitoring and management to clinical risk prediction and healthcare monitoring.
Advances in Time Series Forecasting and Analysis
Sources
Time-Series Foundation Models for ISP Traffic Forecasting
Denoising Refinement Diffusion Models for Simultaneous Generation of Multi-scale Mobile Network Traffic
WaveTuner: Comprehensive Wavelet Subband Tuning for Time Series Forecasting
SimDiff: Simpler Yet Better Diffusion Model for Time Series Point Forecasting
OmniTFT: Omni Target Forecasting for Vital Signs and Laboratory Result Trajectories in Multi Center ICU Data
PeriodNet: Boosting the Potential of Attention Mechanism for Time Series Forecasting
Structured Noise Modeling for Enhanced Time-Series Forecasting
TiCT: A Synthetically Pre-Trained Foundation Model for Time Series Classification
Cisco Time Series Model Technical Report
Zero-Shot Transfer Capabilities of the Sundial Foundation Model for Leaf Area Index Forecasting
RED-F: Reconstruction-Elimination based Dual-stream Contrastive Forecasting for Multivariate Time Series Anomaly Prediction
MSTN: Fast and Efficient Multivariate Time Series Model
The Driver-Blindness Phenomenon: Why Deep Sequence Models Default to Autocorrelation in Blood Glucose Forecasting
Lost in Time? A Meta-Learning Framework for Time-Shift-Tolerant Physiological Signal Transformation
Machine Learning Approaches to Clinical Risk Prediction: Multi-Scale Temporal Alignment in Electronic Health Records