Advances in Time Series Forecasting

The field of time series forecasting is moving towards developing more efficient and effective models that can handle complex dependencies and uncertainties. Recent research has focused on creating lightweight and adaptable models that can learn from multivariate data and capture periodic patterns. Notable advancements include the development of novel neural network architectures and the application of information-theoretic objectives to improve representation learning.

Some notable papers in this area include FaCTR, which proposes a lightweight spatiotemporal Transformer with an explicitly structural design, and TimeMCL, which introduces a method leveraging the Multiple Choice Learning paradigm to forecast multiple plausible time series futures. LightGTS is another noteworthy model, which uses a lightweight general time series forecasting model designed from the perspective of consistent periodical modeling. Additionally, TRACE and STOAT propose innovative approaches to multimodal time series retrieval and spatial-temporal probabilistic causal inference, respectively.

Sources

Winner-takes-all for Multivariate Probabilistic Time Series Forecasting

FaCTR: Factorized Channel-Temporal Representation Transformers for Efficient Time Series Forecasting

LightGTS: A Lightweight General Time Series Forecasting Model

London Blue Light Collaboration Evaluation: A Comparative Analysis of Spatio temporal Patterns on Emergency Services by London Ambulance Service and London Fire Brigade

Diffusion-based Time Series Forecasting for Sewerage Systems

Time Series Representations for Classification Lie Hidden in Pretrained Vision Transformers

Towards Robust Real-World Multivariate Time Series Forecasting: A Unified Framework for Dependency, Asynchrony, and Missingness

InfoDPCCA: Information-Theoretic Dynamic Probabilistic Canonical Correlation Analysis

KARMA: A Multilevel Decomposition Hybrid Mamba Framework for Multivariate Long-Term Time Series Forecasting

Tailored Architectures for Time Series Forecasting: Evaluating Deep Learning Models on Gaussian Process-Generated Data

TRACE: Grounding Time Series in Context for Multimodal Embedding and Retrieval

Multivariate Long-term Time Series Forecasting with Fourier Neural Filter

Age of Information in Unreliable Tandem Queues

Neural Functions for Learning Periodic Signal

STOAT: Spatial-Temporal Probabilistic Causal Inference Network

Time-IMM: A Dataset and Benchmark for Irregular Multimodal Multivariate Time Series