AI Paper Atlas

Research map / Large language models / Agents, code and math

Compositional-ARC: Assessing Systematic Generalization in Abstract Spatial Reasoning

Philipp Mondorf, Shijia Zhou, Monica Riedler, Barbara Plank

ICLR 2026 · Large language models › Agents, code and math · approach: Inductive and abstract reasoning evaluation · 1 citations (OpenAlex)

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 auxiliary copy task to MLC training (ablation)MLC · Compositional-ARC Systematicity (mean of 4 splits)exact match accuracy (%): 69.05 -> 86.73Improved
add primitive transformations to the study examples (ablation)MLC · Compositional-ARC Systematicity (mean of 4 splits)exact match accuracy (%): 75.27 -> 86.73Improved
add level-1 compositions to the study examples (ablation)MLC · Compositional-ARC Systematicity (mean of 4 splits)exact match accuracy (%): 21.01 -> 86.73Improved
add test-time training on study examplesLlama-3.2-3B-ReARC · Compositional-ARC Systematicityexact match accuracy (%): 0.87 -> 73.70Improved
add test-time training on study examplesMistral-NeMO-Minitron-8B-Full · Compositional-ARC Systematicityexact match accuracy (%): 0.70 -> 78.20Improved
add image input to the text promptGPT-4o · Compositional-ARC 3-Shotexact match accuracy (%): 22.28 -> 19.42Hurt
add image input to the text promptGemini 2.0 Flash · Compositional-ARC 3-Shotexact match accuracy (%): 30.08 -> 17.19Hurt
add image input to the text promptGPT-4o · Compositional-ARC Systematicityexact match accuracy (%): 0.99 -> 0.86Hurt
add image input to the text promptGemini 2.0 Flash · Compositional-ARC Systematicityexact match accuracy (%): 2.66 -> 2.05Hurt
switch from 3-shot examples to systematic primitive+level-1 examplesMLC · Compositional-ARCexact match accuracy (%): 99.92 -> 78.26Hurt
expand transformation variety (more translations, extensions, colors)MLC · Compositional-ARC Systematicityexact match accuracy (%): 88.10No clear effect
use MLC small model instead of o3-miniCompositional-ARC Systematicity · Compositional-ARC Systematicityexact match accuracy (%): 0.53 -> 78.26Improved

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