paper-with-me

홈 › Papers

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

2025-09-29 · Lukas Rauch, René Heinrich, Houtan Ghaffari, Lukas Miklautz, Ilyass Moummad, Bernhard Sick, Christoph Scholz arxiv

Although probing frozen models has become a standard evaluation paradigm, self-supervised learning in audio defaults to fine-tuning when pursuing state-of-the-art on AudioSet. A key reason is that global pooling creates an information bottleneck causing linear probes to misrepresent the embedding quality: The $\texttt{cls}$-token discards crucial token information about dispersed, localized events in audio. This weakness is rooted in the mismatch between the pretraining objective (globally) and the downstream task (localized). Across a comprehensive benchmark of 13 datasets and 6 spectrogram-based encoders, we investigate the global pooling bottleneck. We introduce binarized prototypical probes: a lightweight and simple pooling method that learns prototypes to perform class-wise information aggregation. Despite its simplicity, our method notably outperforms linear and attentive probing. Our work establishes probing as a competitive and efficient paradigm for evaluating audio SSL models, challenging the reliance on costly fine-tuning.

📄 PDF Abstract BibTeX arXiv:2509.24901

Code (0)

등록된 구현이 없습니다.

Tasks

Self-Supervised LearningAudio Classification

Similar Papers 제목 키워드 기반

Attention, Please! Revisiting Attentive Probing for Masked Image Modeling

2025-06-11 · Bill Psomas, Dionysis Christopoulos, Eirini Baltzi, Ioannis Kakogeorgiou 외

As fine-tuning (FT) becomes increasingly impractical at scale, probing is emerging as the preferred evaluation protocol for self-supervised learning (SSL). Yet, the standard linear probing (LP) fails to adequately reflec…

BenchmarkingComputational EfficiencySelf-Supervised Learning

Rethinking Patch Dependence for Masked Autoencoders

2024-01-25 · Letian Fu, Long Lian, Renhao Wang, Baifeng Shi 외

In this work, we re-examine inter-patch dependencies in the decoding mechanism of masked autoencoders (MAE). We decompose this decoding mechanism for masked patch reconstruction in MAE into self-attention and cross-atten…

DecoderInstance SegmentationRepresentation LearningSemantic Segmentation

Rethink MAE with Linear Time-Invariant Dynamics

2026-04-29 · Zice Wang arxiv

Standard representation probing for visual models relies on mathematically permutation-invariant operations like Global Average Pooling (GAP) or CLS tokens, treating patch representations as an unstructured bag-of-words.…

Where's the Plan? Locating Latent Planning in Language Models with Lightweight Mechanistic Interventions

2026-05-08 · Nicole Ma, Nick Rui arxiv

We study planning site formation in language models -- where internal representations of structurally-constrained future tokens form during the forward pass, and whether they causally drive generation. Using rhyming-coup…

Tracing Pharmacological Knowledge In Large Language Models

2026-03-03 · Basil Hasan Khwaja, Dylan Chen, Guntas Toor, Anastasiya Kuznetsova arxiv

Large language models (LLMs) have shown strong empirical performance across pharmacology and drug discovery tasks, yet the internal mechanisms by which they encode pharmacological knowledge remain poorly understood. In t…

Drug Discovery