GLIDE: Scalable and Event-Aware Inductive Forecasting over Evolving Logistics Graphs

Impact 6 arXiv cs.AI

Summary

  • Researchers introduced GLIDE, a production-ready framework for scalable and event-aware inductive forecasting over evolving logistics graphs in on-demand food delivery platforms.
  • The logistics domain faces challenges including dynamic topologies, scale-volume heterogeneity with zero-inflated long-tailed volumes, and event-driven disruptions.
  • GLIDE features a Graph-Guided Latent Aggregation mechanism that dynamically projects varying physical subgraphs into a fixed-size, semantically stable latent space.

Why it matters

This research provides a scalable solution for real-time order pooling and demand forecasting in complex, dynamic logistics networks.

Who should care