Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MUDAS: Mote-scale Unsupervised Domain Adaptation in Multi-label Sound Classification

Jihoon Yun, Chengzhang Li, Dhrubojyoti Roy, Anish Arora

TL;DR

The paper tackles domain shifts in urban sound classification under severe resource constraints by proposing MUDAS, a mote-scale unsupervised domain adaptation framework for multi-label sound tasks. MUDAS retrains only the downstream classifier on-device using high-confidence target embeddings and introduces class-specific adaptive thresholds, diversity regularization, and distribution alignment inspired by AdaMatch to handle label overlap and robustness. Key contributions include selective on-device relearning, a data-efficiency strategy for storing high-value embeddings, and a loss design that balances source and target signals for multi-label outputs. Evaluations on the SONYC-UST dataset demonstrate improved target-domain performance over baselines while maintaining feasible memory and compute footprints for IoT edge devices, highlighting practical applicability and potential for federated extensions.

Abstract

Unsupervised Domain Adaptation (UDA) is essential for adapting machine learning models to new, unlabeled environments where data distribution shifts can degrade performance. Existing UDA algorithms are designed for single-label tasks and rely on significant computational resources, limiting their use in multi-label scenarios and in resource-constrained IoT devices. Overcoming these limitations is particularly challenging in contexts such as urban sound classification, where overlapping sounds and varying acoustics require robust, adaptive multi-label capabilities on low-power, on-device systems. To address these limitations, we introduce Mote-scale Unsupervised Domain Adaptation for Sounds (MUDAS), a UDA framework developed for multi-label sound classification in resource-constrained IoT settings. MUDAS efficiently adapts models by selectively retraining the classifier in situ using high-confidence data, minimizing computational and memory requirements to suit on-device deployment. Additionally, MUDAS incorporates class-specific adaptive thresholds to generate reliable pseudo-labels and applies diversity regularization to improve multi-label classification accuracy. In evaluations on the SONYC Urban Sound Tagging (SONYC-UST) dataset recorded at various New York City locations, MUDAS demonstrates notable improvements in classification accuracy over existing UDA algorithms, achieving good performance in a resource-constrained IoT setting.

MUDAS: Mote-scale Unsupervised Domain Adaptation in Multi-label Sound Classification

TL;DR

The paper tackles domain shifts in urban sound classification under severe resource constraints by proposing MUDAS, a mote-scale unsupervised domain adaptation framework for multi-label sound tasks. MUDAS retrains only the downstream classifier on-device using high-confidence target embeddings and introduces class-specific adaptive thresholds, diversity regularization, and distribution alignment inspired by AdaMatch to handle label overlap and robustness. Key contributions include selective on-device relearning, a data-efficiency strategy for storing high-value embeddings, and a loss design that balances source and target signals for multi-label outputs. Evaluations on the SONYC-UST dataset demonstrate improved target-domain performance over baselines while maintaining feasible memory and compute footprints for IoT edge devices, highlighting practical applicability and potential for federated extensions.

Abstract

Unsupervised Domain Adaptation (UDA) is essential for adapting machine learning models to new, unlabeled environments where data distribution shifts can degrade performance. Existing UDA algorithms are designed for single-label tasks and rely on significant computational resources, limiting their use in multi-label scenarios and in resource-constrained IoT devices. Overcoming these limitations is particularly challenging in contexts such as urban sound classification, where overlapping sounds and varying acoustics require robust, adaptive multi-label capabilities on low-power, on-device systems. To address these limitations, we introduce Mote-scale Unsupervised Domain Adaptation for Sounds (MUDAS), a UDA framework developed for multi-label sound classification in resource-constrained IoT settings. MUDAS efficiently adapts models by selectively retraining the classifier in situ using high-confidence data, minimizing computational and memory requirements to suit on-device deployment. Additionally, MUDAS incorporates class-specific adaptive thresholds to generate reliable pseudo-labels and applies diversity regularization to improve multi-label classification accuracy. In evaluations on the SONYC Urban Sound Tagging (SONYC-UST) dataset recorded at various New York City locations, MUDAS demonstrates notable improvements in classification accuracy over existing UDA algorithms, achieving good performance in a resource-constrained IoT setting.

Paper Structure

This paper contains 14 sections, 14 equations, 5 figures, 5 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORKS
  3. Multi-label Sound Classification
  4. Unsupervised Domain Adaptation
  5. On-Device Learning
  6. SYSTEM MODEL & PROBLEM STATEMENT
  7. UNSUPERVISED DOMAIN ADAPTATION IN MULTI-LABEL CLASSIFICATION
  8. Pre-Processing for Acoustic Data
  9. Training Classifier Model with Feature Embeddings
  10. ON-DEVICE RESOURCE-EFFICIENT RELEARNING
  11. Optimization in Relearning Algorithm
  12. Selection for Data Used in Relearning
  13. EVALUATION
  14. CONCLUSIONS

Figures (5)

  • Figure 1: The MUDAS Framework
  • Figure 2: Flow of computation in MUDAS training
  • Figure 3: Micro AUPRC performance of varying unlabeled target dataset sizes vs. consistent dataset size with different splits (3-class evaluation)
  • Figure 4: Impact of varying user-defined parameters for positive and negative thresholds, $\tau^+$ and $\tau^-$ (6-class evaluation)
  • Figure 5: Ablation study: Impact of data selection on MUDAS performance (6-class evaluation)