FAME: Forecastability-Aware Mixture of Experts for Heterogeneous Time Series Forecasting

2026-06-08 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, medium

Summary

FAME, a novel sparse mixture-of-experts framework, addresses the challenge of forecasting heterogeneous time series in large-scale retail and industrial systems. It introduces forecastability-aware expert routing, learning how data characteristics determine the suitability of forecasting experts. The system represents each time series with a multidimensional "forecastability fingerprint," mines expert-suitability targets from validation performance, and trains a cost-aware sparse router to activate a small, budgeted set of experts for each series. Evaluated on a production-scale vending-machine sales dataset from Shandong New Beiyang (SNBC) with over 5,000 machines and 60 million transactions, FAME Top-2 reduced Mean Squared Error (MSE) by 12.4% compared to LightGBM, the strongest single expert, while executing an average of 1.92 experts per series. This framework transforms heterogeneous sales forecasting from heuristic model selection into data mining of forecastability patterns and expert specialization.

Key takeaway

For Machine Learning Engineers managing large-scale time series forecasting systems, FAME offers a robust approach to improve accuracy and efficiency. You should consider implementing forecastability-aware expert routing to dynamically select specialized models for diverse series. This method reduces Mean Squared Error by identifying optimal expert combinations, moving beyond heuristic model selection. Integrate this framework to enhance demand forecasts and optimize inventory planning.

Key insights

FAME uses forecastability fingerprints and a sparse router to dynamically select optimal experts for heterogeneous time series.

Principles

• Heterogeneous series benefit from specialized experts.
• Forecastability patterns guide expert selection.
• Sparse routing reduces inference cost.

Method

FAME represents series with forecastability fingerprints, mines expert-suitability targets from validation, and trains a cost-aware sparse router to activate budgeted experts.

In practice

• Apply forecastability fingerprints to series.
• Train sparse routers for expert selection.
• Integrate into replenishment-planning pipelines.

Topics

• Time Series Forecasting
• Mixture-of-Experts
• Forecastability Fingerprint
• Sparse Routing
• Retail Forecasting
• Demand Forecasting

Code references

• hit636/FAME
• BruceZhangReve/MoME

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• From Long News to Accurate Forecast: Importance-Aware Fusion and PRM-Guided Reflection for Time Series Forecasting — Importance-aware compression and PRM-guided retrieval improve news-enhanced time series forecasting by managing context and selection.
• Discovering Entity-Conditioned Lag Heterogeneity: A Lag-Gated Neural Audit Framework for Panel Time Series — AC-GATE provides a verifiable framework to expose entity-conditioned lag heterogeneity as a structured, inspectable model output.
• #356 The Forecast for Time Series Forecasts with Rami Krispin, Senior Manager of Data Science at Apple — Time series foundation models enable scalable forecasting for massive datasets, moving beyond traditional methods.
• Is Your Time Series Even Forecastable? — Forecastability is a data property, not a model property, determining a time series' theoretical prediction ceiling.
• Vessel Traffic Flow Prediction on Sparse Data via Spatio-Temporal Graph Neural Networks with a Learnable Tweedie Head — A learnable Tweedie head improves vessel traffic prediction on sparse data by capturing heterogeneous variability.
• Ensemble High-Resolution Weather Forecasting on AMD Instinct GPU Accelerators — Diffusion models enable robust ensemble weather forecasting by generating diverse atmospheric states to quantify uncertainty.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.