EXAONE Demand 1.0: A Time Series Foundation Model for Demand Forecasting

2026-09-25

AI Summary

This study addresses the limitations of general-purpose time series foundation models in handling demand data characterized by short histories, frequent zero values, and stockout-induced truncation. We propose a specialized adaptation framework for demand forecasting that introduces a demand-specific corpus synthesis mechanism and designs a low-rank branch adapter employing dynamic routing based on eight-dimensional scale-free statistics. This architecture optimizes predictive performance across four distinct demand patterns while keeping the backbone network frozen. Extensive experiments demonstrate that the proposed method outperforms 36 baseline models across 22 datasets. Furthermore, our results confirm that mixed training with both real and synthetic data yields significantly superior performance compared to purely synthetic approaches.

Abstract

Time series foundation models (TSFMs) are pretrained on series from diverse domains, where demand series make up only a small fraction. Demand data has properties that such corpora rarely contain: Short histories, frequent zeros, censoring by stock-outs, and exogenous events that the series does not record. To this end, we propose EXAONE Demand, built on 1) a demand-specific corpus and 2) a demand-aware adapter. For the corpus, we assemble 11.3M series and 48.4B observations from 73 sources, and a synthetic generator supplies the behaviour that open demand data under-represents. For the adapter, we attach low-rank branches to a frozen general-domain backbone, one for each of the four demand classes (smooth, intermittent, erratic, and lumpy), and a router that reads eight scale-free statistics of the input series decides how much each branch contributes. We build EXAONE Demand in two versions, one trained on real-world and synthetic demand together and one trained on the synthetic corpus alone. On 22 held-out datasets, both versions outperform 36 TSFMs, and real-world demand adds a gain over synthetic data alone.

Problem

• Demand Forecasting
• Time Series Foundation Models
• Intermittent Demand
• Data Censoring
• Exogenous Events

Innovation

• Time Series Foundation Model
• Demand Forecasting
• Low-Rank Adapter
• Synthetic Data Generation
• Mixture of Experts