AI RESEARCH
Versioned Late Materialization for Ultra-Long Sequence Training in Recommendation Systems at Scale
arXiv CS.AI
•
ArXi:2604.24806v1 Announce Type: cross Modern Deep Learning Recommendation Models (DLRMs) follow scaling laws with sequence length, driving the frontier toward ultra-long User Interaction History (UIH). However, the industry-standard "Fat Row" paradigm, which pre-materializes these sequences into every