AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Geometric Constraints for Small Language Models to Understand and Expand Scientific Taxonomies

Liri Fang, Dongqi Fu, Jiawei Han, Jingrui He, Vetle I Torvik

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Calibration and knowledge in LLM tuning

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add hyperbolic structure-dominated encoder (weight 0.5) (ablation)SS-MONO (w/o AD) · SemEval-FoodMRR: 0.399 -> 0.430Improved
use hyperbolic encoder with cone loss instead of Euclidean (ablation)SS-MONO · SemEval-FoodMRR: 0.175 -> 0.400Improved
add sequential self-attention in context encoder (ablation)SS-MONO · SemEval-FoodMRR: 0.065 -> 0.430Improved
use self-attention instead of GAT (ablation)SS-MONO · SemEval-FoodMRR: 0.215 -> 0.430Improved
use self-attention instead of GCN (ablation)SS-MONO · SemEval-FoodMRR: 0.113 -> 0.430Improved
add structure loss on all neighbor types (1111) (ablation)SS-MONO (w/o AD) · SemEval-FoodMRR: 0.217 -> 0.430Improved
sample 3 hops instead of 2 (ablation)SS-MONO · SemEval-FoodMRR: 0.383 -> 0.430Improved
sample 3 hops instead of 4 (ablation)SS-MONO · SemEval-FoodMRR: 0.406 -> 0.430Improved
add LLM augmented edge descriptions (ablation)SS-MONO (w/o AD) · SemEval-FoodMR: 315.79 -> 239.17Improved
add LLM augmented descriptions (ablation)SS-MONO (w/o AD) · WordNet-VerbMRR: 0.297 -> 0.334Improved
add LLM augmented descriptions (ablation)SS-MONO (w/o AD) · SemEval-FoodMRR: 0.430 -> 0.400Hurt
LLM reranking of top-100 candidates (ablation)SS-MONO · SemEval-FoodMRR: 0.400 -> 0.462Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy