AI Paper Atlas

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

Spherical Procrustes Alignment for Reliable Medical Audio Diagnosis

Ying Wang, Guoheng Huang, Chan-Tong Lam, Xiaochen Yuan

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Dense features and foundation priors

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 spherical branch (L2 normalization) (ablation)BEATs · ICBHI 2017 4-class (BEATs)ECE (%): 28.51 -> 19.37Improved
add fixed ETF geometry (ablation)BEATs + Spherical branch · ICBHI 2017 4-class (BEATs)ECE (%) / AS: ECE 19.37 -> 9.37; AS 63.96 -> 62.83Mixed
add dynamic Procrustes alignment (ablation)BEATs + Spherical + ETF · ICBHI 2017 4-class (BEATs)ECE (%) / AS: ECE 9.37 -> 4.87; AS 62.83 -> 63.48Improved
add self-distillation (KL between branches) (ablation)BEATs + Spherical + ETF + Procrustes · ICBHI 2017 4-class (BEATs)ECE (%) / AS: AS 63.48 -> 64.42; ECE 4.87 -> 4.44Improved
add SPABEATs · ICBHI 2017 4-class (BEATs)ECE (%): 28.51 -> 4.44Improved
add SPAPAFA · ICBHI 2017 4-classAverage Score (%): 64.45 -> 65.27Improved
add SPAM2D · CirCor DigiScopeECE (%): 9.17 -> 4.63Improved
use Focal Loss or LDAM-DRW instead of CECNN6 · ICBHI 2017 4-classAverage Score (%): 58.66 -> 57.14 (Focal), 56.93 (LDAM-DRW)Hurt
use MixupBEATs · ICBHI 2017 4-class (BEATs)Average Score (%): 63.90 -> 62.16Hurt
use Label SmoothingBEATs · ICBHI 2017 4-class (BEATs)ECE (%): 28.51 -> 33.91Hurt
use Temperature ScalingBEATs · ICBHI 2017 4-class (BEATs)ECE (%): 28.51 -> 21.99Improved
use MC Dropout or Deep Ensembles (5 passes)BEATs · ICBHI 2017 4-class (BEATs)inference time (ms): 4.1 -> 19.1 (MC Dropout), 20.5 (Deep Ensembles)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