AI Paper Atlas

Research map / Learning theory and algorithms / Combinatorial and search optimization

Error Slice Discovery via Manifold Compactness

Han Yu, Hao Zou, Jiashuo Liu, Renzhe Xu, Yue He, Xingxuan Zhang, Peng Cui

AAAI 2026 · Learning theory and algorithms › Combinatorial and search optimization · approach: Constrained neural network optimization · 4 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
replace Domino/Spotlight/PlaneSpot with MCSD (joint risk and manifold compactness QP)error slice discovery · dcbench (correlation, rare, noisy)precision@10 (%): 36.2 -> 47.4 (correlation, vs Domino)Improved
use MCSD instead of Domino on CelebA blond-hair categoryerror slice discovery · CelebA blond-hair categorymanifold compactness: 6.07 -> 8.09Improved
MCSD vs Spotlight on slice accuracy in CelebA blond-hairerror slice discovery · CelebA blond-hair categoryaccuracy (%): 26.3 (Spotlight) vs 33.8 (MCSD)Hurt
apply MCSD on BDD100K object detection (loss-based, not classification only)error slice discovery · BDD100K pedestrianmanifold compactness: 2.05 -> 6.60Improved
MCSD vs Spotlight on BDD100K traffic-light slice APerror slice discovery · BDD100K traffic lightAP (%): 46.3 (Spotlight) vs 57.3 (MCSD)Hurt
use MCSD on CivilComments toxic categoryerror slice discovery · CivilComments toxicaccuracy (%): 48.6 -> 25.2 (vs Spotlight; lower is better)Improved
measure coherence by manifold compactness rather than variance across slice granularitycoherence metric · CelebAcoherence monotonicityImproved

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