Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper addresses limitations in Deep Feature Synthesis (DFS) for relational learning with tabular foundation models, specifically feature explosion and interaction blindness. The authors introduce RelICL, a novel approach that propagates and fuses information step-by-step through the schema graph using the same foundation model. RelICL achieves performance on par with DFS on the RelBench benchmark while mitigating the identified scalability and interaction issues.
Relational learning with tabular foundation models can now bypass feature explosion and interaction blindness by propagating information directly through the schema graph.
Tabular foundation models achieve state-of-the-art performance on single-table tasks without any training. Recent work suggests that they are also well-suited for relational learning via deep feature synthesis (DFS), which flattens a relational schema into a single table by adding aggregates of the other tables'columns as features. This approach is appealing because it directly benefits from improvements to or customization of the underlying tabular foundation model. In this paper, we identify two key problems with DFS: feature explosion and interaction blindness. The first problem arises because the number of DFS features grows quickly as the schema becomes more complex, limiting scalability and performance. The second problem arises because column-wise aggregates do not account for feature interactions, limiting performance. We propose and explore an alternative method termed RelICL, which keeps the benefits of DFS but alleviates these two problems. At its heart, RelICL propagates and fuses information step by step through the schema graph, using the same tabular foundation model that is eventually used for prediction to do so. In our experimental study using RelBench tasks, RelICL was on par with the strongest approach based on deep feature synthesis.