AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Unmute the Patch Tokens: Rethinking Probing in Multi-Label Audio Classification

Lukas Rauch, René Heinrich, Houtan Ghaffari, Lukas Miklautz, Ilyass Moummad, Bernhard Sick, Christoph Scholz

ICLR 2026 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Self-supervised speech representations · 5 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 the linear cls-token probe with binarized prototypical poolingEAT frozen probe · AudioSet-20kmAP: 17.29 -> 31.67Improved
replace the linear cls-token probe with binarized prototypical poolingA-MAE frozen probe · AudioSet-20kmAP: 8.36 -> 22.32Improved
use protobin instead of linear across general audio datasetsall six backbones · general multi-label audiomAP: +14.41Improved
use protobin instead of linear on few-shot bioacousticsall six backbones · birdset few-shotmAP: +12.16Improved
use prototypical pooling instead of the best attentive pooling (mhca)frozen backbones · general audiomAP: -4.59Improved
replace mhca with protobin on polyphonic AudioSet-20kBEATs · AudioSet-20kmAP: 21.86 -> 31.54Improved
replace mhca with protobin on single-label ESC-50EAT · ESC-50 (accuracy)mAP: 89.8 -> 86.8Hurt
replace mhca with protobin on single-label SC-2EAT · SC-2 (accuracy)mAP: 93.2 -> 90.4Hurt
use protobin with ASiT backbone vs float class-dependent protoASiT · AudioSet-20kmAP: 21.89 -> 20.96Hurt
add supervised AudioSet fine-tuning of the backbone before linear probingBEATs linear probe · AudioSet-20kmAP: 24.71 -> 40.30Improved
add supervised fine-tuning to the backbone before protobin probingSSLAM protobin · AudioSet-20kmAP: 30.94 -> 43.70Improved
use protobin on top of linear probe so as to close the gap to fine-tuningEAT · AudioSet-20kmAP: closes 63% of the gapImproved

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