Context Window Failures in Relational Foundation Models
Recent Relational Deep Learning architectures have been proposed as foundation models for multi-table relational data, yet they impose constrained neighborhood budgets that force row truncation when an entity has many related records. We introduce Animus, a synthetic financial dataset in which predicting customer incom...
D. Correa, Francisco Galuppo Azevedo
· 0 citations