Paper accepted at COLM oral: richer graph representations can generalise worse
PMinervini · x · 2026-10-21
Alessio Cagiano and colleagues' paper "Graph-to-Table Interfaces Control Relational Generalisation" is accepted as an oral at COLM.
The work asks a counterintuitive question: can a richer graph representation generalise worse? By casting relational reasoning as a graph-to-table interface, the authors systematically test how changing the row encoder, the labelled context, or the decoder affects relational generalisation, with the setup illustrated via hypernym queries like wolf → canine.
More from Research
- Hugging Face Launches SimReady Central: Open Sim-Ready 3D Assets for Robotics — Hugging Face Blog · 2026-10-21
- MARTINI algorithm mixes surrogate rewards to steer generative models without querying the oracle — mmbronstein · 2026-10-21
- GPT Image 2.5 vs Nano Banana 2.1: 502 synthetic rust masks reviewed, 79.7% retained — Gloomy_Recognition_4 · 2026-10-21
- NVIDIA cuDSS hits GA: GPU sparse solvers deliver up to 10x speedups in engineering simulation tools — NVIDIA Blog · 2026-10-21
- Chaperone design agent cuts token use 86%, yields 1.32 nM cortisol binder in two weeks — josephdviviano · 2026-10-21
- Mathematicians clash: do self-maintaining biological neurons make AI fundamentally different? — ValerioCapraro · 2026-10-21