Ensemble of pruned models for low-complexity acoustic scene classification
For the DCASE 2020 Challenge, the focus of Task 1B is to develop low-complexity models for classification of 3 different types of acoustic scenes, which have potential applications in resource-scarce edge devices deployed in a large-scale acoustic network. In this paper, we present the training meth...
Saved in:
Main Authors: | , , |
---|---|
Other Authors: | |
Format: | Conference or Workshop Item |
Language: | English |
Published: |
2021
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/148327 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
id |
sg-ntu-dr.10356-148327 |
---|---|
record_format |
dspace |
spelling |
sg-ntu-dr.10356-1483272021-05-10T00:57:38Z Ensemble of pruned models for low-complexity acoustic scene classification Ooi, Kenneth Peksi, Santi Gan, Woon-Seng School of Electrical and Electronic Engineering 5th Workshop on Detection and Classification of Acoustic Scenes and Events, DCASE 2020 Digital Signal Processing Laboratory Engineering::Computer science and engineering Acoustic Scene Classification Deep Learning For the DCASE 2020 Challenge, the focus of Task 1B is to develop low-complexity models for classification of 3 different types of acoustic scenes, which have potential applications in resource-scarce edge devices deployed in a large-scale acoustic network. In this paper, we present the training methodology for our submissions for the challenge, with the best-performing system consisting of an ensemble of VGGNet- and Inception-Net-based lightweight classification models. The subsystems in the ensemble classifier were pruned by setting low-magnitude weights periodically to zero with a polynomial decay schedule to achieve an 80% reduction in individual subsystem size. The resultant ensemble classifier outperformed the baseline model on the validation set over 10 runs and had 119758 non-zero parameters taking up 468KB of memory. This shows the efficacy of the pruning technique used. We also performed experiments to compare the performance of various data augmentation schemes, input feature representations, and model architectures in our training methodology. No external data was used, and source code for the submission can be found at https://github.com/kenowr/DCASE-2020-Task-1B. Ministry of Education (MOE) Accepted version Supported by the Singapore Ministry of Education Academic Research Fund Tier-2, under research grant MOE2017-T2-2-060. 2021-05-10T00:57:38Z 2021-05-10T00:57:38Z 2020 Conference Paper Ooi, K., Peksi, S. & Gan, W. (2020). Ensemble of pruned models for low-complexity acoustic scene classification. 5th Workshop on Detection and Classification of Acoustic Scenes and Events, DCASE 2020. https://hdl.handle.net/10356/148327 en MOE2017-T2-2-060 © 2020 DCASE. All rights reserved. application/pdf |
institution |
Nanyang Technological University |
building |
NTU Library |
continent |
Asia |
country |
Singapore Singapore |
content_provider |
NTU Library |
collection |
DR-NTU |
language |
English |
topic |
Engineering::Computer science and engineering Acoustic Scene Classification Deep Learning |
spellingShingle |
Engineering::Computer science and engineering Acoustic Scene Classification Deep Learning Ooi, Kenneth Peksi, Santi Gan, Woon-Seng Ensemble of pruned models for low-complexity acoustic scene classification |
description |
For the DCASE 2020 Challenge, the focus of Task 1B is to develop low-complexity models for classification of 3 different types of acoustic scenes, which have potential applications in resource-scarce edge devices deployed in a large-scale acoustic network. In this paper, we present the training methodology for our submissions for the challenge, with the best-performing system consisting of an ensemble of VGGNet- and Inception-Net-based lightweight classification models. The subsystems in the ensemble classifier were pruned by setting low-magnitude weights periodically to zero with a polynomial decay schedule to achieve an 80% reduction in individual subsystem size. The resultant ensemble classifier outperformed the baseline model on the validation set over 10 runs and had 119758 non-zero parameters taking up 468KB of memory. This shows the efficacy of the pruning technique used. We also performed experiments to compare the performance of various data augmentation schemes, input feature representations, and model architectures in our training methodology. No external data was used, and source code for the submission can be found at https://github.com/kenowr/DCASE-2020-Task-1B. |
author2 |
School of Electrical and Electronic Engineering |
author_facet |
School of Electrical and Electronic Engineering Ooi, Kenneth Peksi, Santi Gan, Woon-Seng |
format |
Conference or Workshop Item |
author |
Ooi, Kenneth Peksi, Santi Gan, Woon-Seng |
author_sort |
Ooi, Kenneth |
title |
Ensemble of pruned models for low-complexity acoustic scene classification |
title_short |
Ensemble of pruned models for low-complexity acoustic scene classification |
title_full |
Ensemble of pruned models for low-complexity acoustic scene classification |
title_fullStr |
Ensemble of pruned models for low-complexity acoustic scene classification |
title_full_unstemmed |
Ensemble of pruned models for low-complexity acoustic scene classification |
title_sort |
ensemble of pruned models for low-complexity acoustic scene classification |
publishDate |
2021 |
url |
https://hdl.handle.net/10356/148327 |
_version_ |
1701270615223173120 |