AI Paper Atlas

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

Richer Representations for Neural Algorithmic Reasoning via Auxiliary Reconstruction

Jiafu Huang, Chao Peng, Chenyang Xu, Zhengfeng Yang, Kecheng Cai, Chenhao Zhang, Yi Wang, Yiwei Gong, Wanqin Zhou, Irene Zheng

AAAI 2026 · Large language models › Agents, code and math · approach: Inductive and abstract reasoning evaluation · 0 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 reconstruction loss and GNN encoder (ReNAR)Triplet-GMPNN · CLRS overall averageF1 accuracy (CLRS): 83.63% -> 87.11%Improved
add hint-level masking to reconstruction (M-ReNAR) (ablation)ReNAR · CLRS overall averageF1 accuracy (CLRS): 87.11% -> 88.41%Improved
add reconstruction frameworkTriplet-GMPNN · CLRS GraphsF1 accuracy (CLRS): 88.80% -> 94.74%Improved
add reconstruction frameworkTriplet-GMPNN · CLRS DFSF1 accuracy (CLRS): around 60% -> nearly 100%Improved
add hint-level masking (ablation)ReNAR · CLRS StringsF1 accuracy (CLRS): 62.70% -> 74.72%Improved
add hint-level masking (ablation)ReNAR · CLRS SortingF1 accuracy (CLRS): 92.13% -> 90.51%Hurt
replace GNN+gated encoder with linear encoder (ReNAR-linear) (ablation)ReNAR · CLRS (Triplet-GMPNN)F1 accuracy (CLRS)Hurt
vary masking ratio beta in {0.1,0.3,0.5,0.7,0.9} (ablation)M-ReNAR · CLRS categoriesF1 accuracy (CLRS)Mixed
vary reconstruction weight lambda in {0.0,0.05,0.1,0.15,0.20} (ablation)ReNAR · CLRS categoriesF1 accuracy (CLRS)Mixed
add ReNAR to other processors (Triplet-PGN, Triplet-MPNN)Triplet-PGN / Triplet-MPNN · CLRS overall averageF1 accuracy (CLRS)Improved
add reconstruction frameworkTriplet-GMPNN · CLRS SearchF1 accuracy (CLRS): 61.14% -> 58.89%Hurt

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