---
title: "Bridging Neural Representations and Symbolic Semantics in Audio Classification"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.40414v1"
permalink: "https://feed.astralor.com/items/itm_8ee4f6f344929bef0e3c118b"
published: "2026-10-29T16:42:13.000Z"
score: 54
channel: "tech"
tags: ["多模态","Bridging Neural Representations","Symbolic Semantics","Audio Classification Deep","Existing","Deep","Interpreting"]
---

# Bridging Neural Representations and Symbolic Semantics in Audio Classification

## 速览

Deep audio classifiers can achieve strong predictive performance, but their hidden representations remain difficult to interpret. Existing audio explainability methods often focus on input-level attribution, such as wav…

## 关键点

- Deep audio classifiers can achieve strong predictive performance, but their hidden representations remain difficult to interpret.
- Existing audio explainability methods often focus on input-level attribution, such as waveform segments, spectrogram regions, or acoustic features, and provide limited insight into what hidden neurons encode.
- Interpreting hidden neurons at the concept level can provide a more meaningful view of what an audio model learns internally.

## 全文

Deep audio classifiers can achieve strong predictive performance, but their hidden representations remain difficult to interpret. Existing audio explainability methods often focus on input-level attribution, such as waveform segments, spectrogram regions, or acoustic features, and provide limited insight into what hidden neurons encode. Interpreting hidden neurons at the concept level can provide a more meaningful view of what an audio model learns internally. This paper investigates whether hidden neurons in audio classifiers can be associated with human-understandable symbolic acoustic concepts. A model-agnostic neurosymbolic framework is presented for analyzing the final hidden layer of audio classifiers, containing 32 neurons, through symbolic concept induction. The framework is evaluated on two audio classification settings: music genre classification using GTZAN dataset and musical instrument classification using IRMAS dataset. Results show that several hidden neurons can be associated with meaningful induced concepts, and most confirmed neuron-concept pairs are statistically supported using the Mann-Whitney U test. These findings suggest that symbolic concept induction is a promising direction for explaining latent audio representations beyond input-level attribution.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_8ee4f6f344929bef0e3c118b