AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

The Natural Geometry of Code: Hyperbolic Representation Learning for Program Reasoning

Weilin Zhou

ICLR 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive representation theory

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
use hyperbolic (Lorentz) operations instead of Euclidean counterparts (ablation)HypeCodeNet · BigCloneBenchF1-score: 0.923 -> 0.940Improved
add geodesic-based (hyperbolic) attention (ablation)HypeCodeNet · BigCloneBenchF1-score: 0.932 -> 0.940Improved
add curvature annealing (ablation)HypeCodeNet · BigCloneBenchF1-score: 0.935 -> 0.940Improved
increase embedding dimension from 32 to 768 (ablation)HypeCodeNet · BigCloneBenchF1-score: 0.928 -> 0.940Improved
increase embedding dimension from 64 to 128 (ablation)HypeCodeNet · BigCloneBenchF1-score: 0.934 -> 0.938Improved
increase embedding dimension from 128 to 256 (ablation)HypeCodeNet · BigCloneBenchF1-score: 0.938 -> 0.939No clear effect
hyperbolic GNN instead of best Euclidean hybrid baselineHypeCodeNet vs CodeFORMER · BigCloneBenchF1-score: 0.928 -> 0.940Improved
hyperbolic GNN instead of best Euclidean hybrid baselineHypeCodeNet vs CodeFORMER · POJ-104accuracy: 0.974 -> 0.981Improved
hyperbolic GNN instead of best Euclidean hybrid baselineHypeCodeNet vs CodeFORMER · CodeXGLUE completion (Python)accuracy (%): 44.0 -> 45.0Improved
hyperbolic GNN instead of best Euclidean hybrid baselineHypeCodeNet vs CodeFORMER · CodeXGLUE completion (Java)accuracy (%): 41.4 -> 42.2Improved
hyperbolic GNN instead of best Euclidean hybrid baselineHypeCodeNet vs CodeFORMER · GitHub Java call-graph link predictionAUC: 0.915 -> 0.965Improved
hyperbolic GNN instead of best Euclidean hybrid baselineHypeCodeNet vs CodeFORMER · GitHub Java call-graph link predictionHits@10: 0.768 -> 0.820Improved

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