KAIROS: Unified Training for Universal Non-Autoregressive Time Series Forecasting

2025-10-02

AI Summary

Real-time time-series forecasting for web applications demands both low latency and high fidelity; however, autoregressive methods suffer from error accumulation, while existing non-autoregressive approaches often produce over-smoothed predictions. To address this, we propose a unified non-autoregressive modeling paradigm that abandons point-wise generation and instead directly models segment-level multimodal probability distributions—enabling full-horizon prediction in a single forward pass. Our method is trained end-to-end on large-scale time-series corpora and integrates piecewise temporal modeling with multimodal density estimation. Evaluated on six benchmarks, it achieves zero-shot generalization performance competitive with state-of-the-art large time-series models, reduces inference latency by an order of magnitude, and cuts GPU memory consumption by over 40%. This work constitutes the first demonstration of scalability and practical viability of non-autoregressive modeling for general-purpose time-series forecasting.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Natural Language Processing: (Large) Language Models
• Computer Vision: Multi-modal Vision

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

In the World Wide Web, reliable time series forecasts provide the forward-looking signals that drive resource planning, cache placement, and anomaly response, enabling platforms to operate efficiently as user behavior and content distributions evolve. Compared with other domains, time series forecasting for Web applications requires much faster responsiveness to support real-time decision making. We present KAIROS, a non-autoregressive time series forecasting framework that directly models segment-level multi-peak distributions. Unlike autoregressive approaches, KAIROS avoids error accumulation and achieves just-in-time inference, while improving over existing non-autoregressive models that collapse to over-smoothed predictions. Trained on the large-scale corpus, KAIROS demonstrates strong zero-shot generalization on six widely used benchmarks, delivering forecasting performance comparable to state-of-the-art foundation models with similar scale, at a fraction of their inference cost. Beyond empirical results, KAIROS highlights the importance of non-autoregressive design as a scalable paradigm for foundation models in time series.

Problem

• Developing non-autoregressive forecasting for real-time web applications
• Modeling segment-level multi-peak distributions to prevent over-smoothing
• Achieving zero-shot generalization with reduced inference cost

Innovation

• Non-autoregressive framework for time series forecasting
• Directly models segment-level multi-peak distributions
• Achieves just-in-time inference with zero-shot generalization