Support-Set Target Leakage in Relational Foundation Models during In-Context Learning: Impact, Detection, and Mitigation
This work evaluates a frozen relational encoder with an ICL head on held-out RelBench databases and uses Integrated Gradients to rank and remove suspicious columns, showing that the effect of support-set leakage varies across tasks and relational distances.
Roshan Reddy Upendra, Alexandre Dorais, Joe Meyer et al.
· 0 citations