Time-VLM: Exploring Multimodal Vision-Language Models for Augmented Time Series Forecasting

TL;DR

Time-VLM tackles the challenge of forecasting by fusing temporal dynamics with semantic and visual context through a triple-modal framework. It introduces three dedicated learners—RAL for temporal memory, VAL for time-series-to-image encoding, and TAL for contextual text—operating with frozen Vision-Language Models and a cross-modal fusion module to predict future values. The approach delivers strong few-shot and zero-shot performance, competitive short- and long-term results, and a favorable efficiency profile compared with LLM-based alternatives. This work demonstrates the viability of cross-modal priors to enhance time-series forecasting and outlines directions for domain-specific multimodal pretraining.

Abstract

Recent advancements in time series forecasting have explored augmenting models with text or vision modalities to improve accuracy. While text provides contextual understanding, it often lacks fine-grained temporal details. Conversely, vision captures intricate temporal patterns but lacks semantic context, limiting the complementary potential of these modalities. To address this, we propose \method, a novel multimodal framework that leverages pre-trained Vision-Language Models (VLMs) to bridge temporal, visual, and textual modalities for enhanced forecasting. Our framework comprises three key components: (1) a Retrieval-Augmented Learner, which extracts enriched temporal features through memory bank interactions; (2) a Vision-Augmented Learner, which encodes time series as informative images; and (3) a Text-Augmented Learner, which generates contextual textual descriptions. These components collaborate with frozen pre-trained VLMs to produce multimodal embeddings, which are then fused with temporal features for final prediction. Extensive experiments demonstrate that Time-VLM achieves superior performance, particularly in few-shot and zero-shot scenarios, thereby establishing a new direction for multimodal time series forecasting. Code is available at https://github.com/CityMind-Lab/ICML25-TimeVLM.

Figures (10)

• Figure 1: Our Time-VLM combines text (Right) and vision (Left) modalities to augment time series forecasting.
• Figure 2: Overview of the Time-VLM framework.
• Figure 3: UMAP visualization (left) and gate weight distributions (right) of of multimodal and temporal embeddings.
• Figure 4: Interpretability visualization of Time-VLM: multimodal feature alignment via UMAP.
• Figure 5: Hyperparameters sensitivity analysis on input length, normalization constant, dimension of model and dimension of gate network, reflected by MAE.