ReDiTT: Retrieval Augmented Conditional Diffusion Transformers for Asynchronous Time Series

2026-07-14

AI Summary

This work addresses the joint prediction of the next event time and type in asynchronous time series by proposing a retrieval-augmented conditional diffusion Transformer. The method introduces a retrieval mechanism into diffusion models for the first time, dynamically querying a memory bank in latent space to retrieve historically observed sequences with structural similarity. These retrieved sequences are incorporated via cross-attention as global structural priors to guide the generative process. The proposed framework significantly enhances the stability of long-horizon predictions and improves sample diversity, achieving state-of-the-art performance across seven real-world datasets.

Abstract

We present a diffusion based model for asynchronous time series prediction, where the goal is to predict the next inter event time and event type. To address the inherent uncertainty of future events, we introduce ReDiTT, a retrieval augmented conditional diffusion transformer that operates in latent space. ReDiTT retrieves structurally similar latent sequences from a memory bank during both training and inference and incorporates them as reference conditions through cross attention. This retrieval based conditioning allows the model to attend to relevant temporal dynamics and provides global structural guidance for generation. As a result, ReDiTT stabilizes long horizon forecasting and improves sample diversity. Experiments on seven real world datasets demonstrate state of the art performance on next event prediction and long horizon forecasting. Our code is available at https://github.com/BorealisAI/ReDiTT.

Problem

• asynchronous time series
• event prediction
• inter-event time
• time series forecasting

Innovation

• retrieval-augmented
• conditional diffusion
• asynchronous time series
• latent space
• cross-attention