Image Synthesis as an Intermediate for Controllable Time Series Generation

2026-10-04

AI Summary

This study addresses the limitation that language models directly generating numerical time series often disrupt temporal structures, thereby constraining few-shot forecasting performance. To overcome this, we propose VisualBridge, a novel framework that pioneers the use of time series plots as visual intermediaries to bridge semantic and numerical modalities. Specifically, the method leverages multimodal large language models to extract semantic features and formulate editing strategies, while integrating downstream reward-driven reinforcement learning with a variational autoencoder to achieve controllable sequence generation. This work makes a pioneering contribution by introducing an image-based intermediate representation that effectively fuses semantic and numerical information. Extensive experiments demonstrate that VisualBridge significantly outperforms conventional data augmentation methods on standard benchmarks, with comprehensive ablation studies validating the effectiveness of each individual module.

Abstract

Semantic-driven time-series generation offers a promising way to improve downstream learning in few-shot forecasting, but directly generating numerical sequences from language often fails to preserve the intended temporal structure. We propose VisualBridge, which uses time-series plots as a visual intermediate to bridge high-level temporal semantics and numerical sequences. An MLLM first converts plotted series into structured semantic representations, enabling explicit control over temporal properties such as trend, seasonality, and volatility. We then learn a semantic editing policy with downstream forecasting rewards, allowing the generation process to favor temporal patterns that are beneficial for the target task. The resulting sequences are further modeled by a temporal VAE to produce consistent multivariate augmentations. Experiments on standard public forecasting benchmarks demonstrate that VisualBridge improves few-shot forecasting over conventional augmentation methods, with ablations validating the roles of visual semantic grounding, learned semantic control, and VAE-based generation.

Problem

• Time Series Generation
• Few-shot Forecasting
• Semantic-driven Generation
• Temporal Structure Preservation

Innovation

• Time Series Generation
• Multimodal Large Language Model
• Visual Intermediate Representation
• Semantic Editing Policy
• Temporal VAE