Time Weaver: A Conditional Time Series Generation Model

2024-03-05

International Conference on Machine Learning

AI Summary

Existing time-series generation methods largely neglect conditional modeling of heterogeneous contextual metadata (e.g., weather, location, device status) and fail to adapt well-established conditional generation paradigms from image/audio domains; moreover, no quantitative metric exists to evaluate metadata feature reconstruction fidelity. This paper proposes the first diffusion-based conditional generation framework supporting categorical, continuous, and time-varying metadata. It enables fine-grained conditional control via metadata-aware embedding encoding, conditional feature alignment, and dynamic temporal denoising. We further introduce the first evaluation metric jointly measuring conditional specificity and temporal realism. Extensive experiments across four real-world domains—energy, healthcare, air quality, and transportation—demonstrate that downstream classification tasks achieve up to 27% performance gain over state-of-the-art GAN-based methods.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Time-Series/Data Streams
• Computer Vision: Diffusion Models for Vision

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Imagine generating a city's electricity demand pattern based on weather, the presence of an electric vehicle, and location, which could be used for capacity planning during a winter freeze. Such real-world time series are often enriched with paired heterogeneous contextual metadata (weather, location, etc.). Current approaches to time series generation often ignore this paired metadata, and its heterogeneity poses several practical challenges in adapting existing conditional generation approaches from the image, audio, and video domains to the time series domain. To address this gap, we introduce Time Weaver, a novel diffusion-based model that leverages the heterogeneous metadata in the form of categorical, continuous, and even time-variant variables to significantly improve time series generation. Additionally, we show that naive extensions of standard evaluation metrics from the image to the time series domain are insufficient. These metrics do not penalize conditional generation approaches for their poor specificity in reproducing the metadata-specific features in the generated time series. Thus, we innovate a novel evaluation metric that accurately captures the specificity of conditional generation and the realism of the generated time series. We show that Time Weaver outperforms state-of-the-art benchmarks, such as Generative Adversarial Networks (GANs), by up to 27% in downstream classification tasks on real-world energy, medical, air quality, and traffic data sets.

Problem

• Generating time series using heterogeneous contextual metadata
• Addressing limitations in current conditional generation approaches
• Developing improved evaluation metrics for conditional time series generation

Innovation

• Diffusion model uses heterogeneous metadata for generation
• Novel metric evaluates conditional generation specificity
• Outperforms GANs by 30% in classification tasks