Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AdaProj: Adaptively Scaled Angular Margin Subspace Projections for Anomalous Sound Detection with Auxiliary Classification Tasks

Kevin Wilkinghoff

TL;DR

This work tackles semi-supervised anomalous sound detection by learning embeddings through auxiliary classification tasks. It introduces AdaProj, an angular-margin loss that projects data onto class-specific subspaces, enlarging the optimal solution space and allowing richer normal-data distributions. Empirical results on DCASE2022 and DCASE2023 ASD datasets show AdaProj consistently outperforming existing losses, with notable gains on the more challenging DCASE2023 task. The method promises improved robustness to domain shifts and suggests potential extensions with self-supervised or multi-task learning for broader applicability.

Abstract

The state-of-the-art approach for semi-supervised anomalous sound detection is to first learn an embedding space by using auxiliary classification tasks based on meta information or self-supervised learning and then estimate the distribution of normal data. In this work, AdaProj a novel loss function for training the embedding model is presented. In contrast to commonly used angular margin losses, which project data of each class as close as possible to their corresponding class centers, AdaProj learns to project data onto class-specific subspaces while still ensuring an angular margin between classes. By doing so, the resulting distributions of the embeddings belonging to normal data are not required to be as restrictive as other loss functions allowing a more detailed view on the data. In experiments conducted on the DCASE2022 and DCASE2023 anomalous sound detection datasets, it is shown that using AdaProj to learn an embedding space significantly outperforms other commonly used loss functions.

AdaProj: Adaptively Scaled Angular Margin Subspace Projections for Anomalous Sound Detection with Auxiliary Classification Tasks

TL;DR

This work tackles semi-supervised anomalous sound detection by learning embeddings through auxiliary classification tasks. It introduces AdaProj, an angular-margin loss that projects data onto class-specific subspaces, enlarging the optimal solution space and allowing richer normal-data distributions. Empirical results on DCASE2022 and DCASE2023 ASD datasets show AdaProj consistently outperforming existing losses, with notable gains on the more challenging DCASE2023 task. The method promises improved robustness to domain shifts and suggests potential extensions with self-supervised or multi-task learning for broader applicability.

Abstract

The state-of-the-art approach for semi-supervised anomalous sound detection is to first learn an embedding space by using auxiliary classification tasks based on meta information or self-supervised learning and then estimate the distribution of normal data. In this work, AdaProj a novel loss function for training the embedding model is presented. In contrast to commonly used angular margin losses, which project data of each class as close as possible to their corresponding class centers, AdaProj learns to project data onto class-specific subspaces while still ensuring an angular margin between classes. By doing so, the resulting distributions of the embeddings belonging to normal data are not required to be as restrictive as other loss functions allowing a more detailed view on the data. In experiments conducted on the DCASE2022 and DCASE2023 anomalous sound detection datasets, it is shown that using AdaProj to learn an embedding space significantly outperforms other commonly used loss functions.
Paper Structure (11 sections, 1 theorem, 5 equations, 2 figures, 1 table)

This paper contains 11 sections, 1 theorem, 5 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Notation
  5. AdaProj loss function
  6. Experimental results
  7. Datasets and performance metrics
  8. Anomalous sound detection system
  9. Performance evaluation
  10. Investigating the impact of the subspace dimension on the performance
  11. Conclusions

Key Result

Lemma 2

Let $x\in\mathbb{R}^D$ and let $\mathcal{C}\subset\mathbb{R}^D$ contain pairwise orthonormal elements. If $x\in\mathop{\mathrm{span}}\nolimits(\mathcal{C})\cap \mathcal{S}^{D-1}$, then

Figures (2)

  • Figure 1: Structure of the asd system, adapted from Figure 1 in wilkinghoff2023design. Representation size in each step is given in brackets.
  • Figure 2: Domain-independent performance obtained on the DCASE2023 dataset with different subspace dimensions. The means over ten independent trials are shown.

Theorems & Definitions (5)

  • Definition 1: AdaProj loss
  • Remark
  • Lemma 2
  • proof
  • Remark