Task-Agnostic Structured Pruning of Speech Representation Models

被引:1
|
作者
Wang, Haoyu [1 ]
Wang, Siyuan [1 ]
Zhang, Wei-Qiang [1 ]
Suo, Hongbin [2 ]
Wan, Yulong [2 ]
机构
[1] Tsinghua Univ, Dept Elect Engn, Beijing 100084, Peoples R China
[2] OPPO, Data & AI Engn Syst, Beijing 100026, Peoples R China
来源
基金
中国国家自然科学基金;
关键词
Model pruning; knowledge distillation; model compression; representation learning;
D O I
10.21437/Interspeech.2023-1442
中图分类号
O42 [声学];
学科分类号
070206 ; 082403 ;
摘要
Self-supervised pre-trained models such as Wav2vec2, Hubert, and WavLM have been shown to significantly improve many speech tasks. However, their large memory and strong computational requirements hinder their industrial applicability. Structured pruning is a hardware-friendly model compression technique but usually results in a larger loss of accuracy. In this paper, we propose a fine-grained attention head pruning method to compensate for the performance degradation. In addition, we also introduce the straight through estimator into the L0 regularization to further accelerate the pruned model. Experiments on the SUPERB benchmark show that our model can achieve comparable performance to the dense model in multiple tasks and outperforms the Wav2vec 2.0 base model on average, with 72% fewer parameters and 2 times faster inference speed.
引用
收藏
页码:231 / 235
页数:5
相关论文
共 50 条
  • [31] Latent Plans for Task-Agnostic Offline Reinforcement Learning
    Rosete-Beas, Erick
    Mees, Oier
    Kalweit, Gabriel
    Boedecker, Joschka
    Burgard, Wolfram
    CONFERENCE ON ROBOT LEARNING, VOL 205, 2022, 205 : 1838 - 1849
  • [32] Task-Agnostic Dynamics Priors for Deep Reinforcement Learning
    Du, Yilun
    Narasimhan, Karthik
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 97, 2019, 97
  • [33] Task-agnostic feature extractors for incremental learning at the edge
    Loomis, Lisa
    Wise, David
    Inkawhich, Nathan
    Thiem, Clare
    McDonald, Nathan
    ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING FOR MULTI-DOMAIN OPERATIONS APPLICATIONS VI, 2024, 13051
  • [34] Task-Agnostic Amortized Inference of Gaussian Process Hyperparameters
    Liu, Sulin
    Sun, Xingyuan
    Ramadge, Peter J.
    Adams, Ryan P.
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 33, NEURIPS 2020, 2020, 33
  • [35] TADA: Efficient Task-Agnostic Domain Adaptation for Transformers
    Hung, Chia-Chien
    Lange, Lukas
    Stroetgen, Jannik
    FINDINGS OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS, ACL 2023, 2023, : 487 - 503
  • [36] COSMIC: Mutual Information for Task-Agnostic Summarization Evaluation
    Darrin, Maxime
    Formont, Philippe
    CilEuNG, Jackie Chi Kit
    Piantanida, Pablo
    PROCEEDINGS OF THE 62ND ANNUAL MEETING OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS, VOL 1: LONG PAPERS, 2024, : 12696 - 12717
  • [37] Few-shot Task-agnostic Neural Architecture Search for Distilling Large Language Models
    Xu, Dongkuan
    Mukherjee, Subhabrata
    Liu, Xiaodong
    Dey, Debadeepta
    Wang, Wenhui
    Zhang, Xiang
    Awadallah, Ahmed Hassan
    Gao, Jianfeng
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 35 (NEURIPS 2022), 2022,
  • [38] CEM: Constrained Entropy Maximization for Task-Agnostic Safe Exploration
    Yang, Qisong
    Spaan, Matthijs T. J.
    THIRTY-SEVENTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, VOL 37 NO 9, 2023, : 10798 - 10806
  • [39] Task-Agnostic Continual Hippocampus Segmentation for Smooth Population Shifts
    Gonzalez, Camila
    Ranem, Amin
    Othman, Ahmed
    Mukhopadhyay, Anirban
    DOMAIN ADAPTATION AND REPRESENTATION TRANSFER (DART 2022), 2022, 13542 : 108 - 118
  • [40] FADE: Fusing the Assets of Decoder and Encoder for Task-Agnostic Upsampling
    Lu, Hao
    Liu, Wenze
    Fu, Hongtao
    Cao, Zhiguo
    COMPUTER VISION - ECCV 2022, PT XXVII, 2022, 13687 : 231 - 247