The field of time series forecasting is moving towards incorporating multimodal context and leveraging pre-trained foundation models to enhance predictive accuracy. Researchers are exploring the potential of integrating time series data with other modalities such as text and vision to improve forecasting performance. Notable papers in this area include: UniCast, which introduces a novel parameter-efficient multimodal framework for time series forecasting. EventTSF, which proposes an autoregressive generation framework that integrates historical time series with textual events to make subsequent forecasts. PENGUIN, which enhances Transformer with Periodic-Nested Group Attention for long-term time series forecasting.
Time Series Forecasting Developments
Sources
A Feasibility Experiment on the Application of Predictive Coding to Instant Messaging Corpora
Comparative Analysis of Time Series Foundation Models for Demographic Forecasting: Enhancing Predictive Accuracy in US Population Dynamics
UniCast: A Unified Multimodal Prompting Framework for Time Series Forecasting
CC-Time: Cross-Model and Cross-Modality Time Series Forecasting
Deep Learning-Based Financial Time Series Forecasting via Sliding Window and Variational Mode Decomposition
Towards Unified Multimodal Financial Forecasting: Integrating Sentiment Embeddings and Market Indicators via Cross-Modal Attention
EventTSF: Event-Aware Non-Stationary Time Series Forecasting
PENGUIN: Enhancing Transformer with Periodic-Nested Group Attention for Long-term Time Series Forecasting
Enhancing Forecasting with a 2D Time Series Approach for Cohort-Based Data