Task-specific Compression for Multi-task Language Models using Attribution-based Pruning

被引:0
|
作者
Yang, Nakyeong [1 ]
Jang, Yunah [1 ]
Lee, Hwanhee [2 ]
Jung, Seohyeong [3 ]
Jung, Kyomin [1 ]
机构
[1] Seoul Natl Univ, Seoul, South Korea
[2] Chung Ang Univ, Seoul, South Korea
[3] Hyundai Motor Grp & 42dot Inc, Seoul, South Korea
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Multi-task language models show outstanding performance for various natural language understanding tasks with only a single model. However, these language models utilize an unnecessarily large number of model parameters, even when used only for a specific task. This paper proposes a novel training-free compression method for multi-task language models using a pruning method. Specifically, we use an attribution method to determine which neurons are essential for performing a specific task. We task-specifically prune unimportant neurons and leave only task-specific parameters. Furthermore, we extend our method to be applicable in low-resource and unsupervised settings. Since our compression method is training-free, it uses few computing resources and does not destroy the pre-trained knowledge of language models. Experimental results on the six widely-used datasets show that our proposed pruning method significantly outperforms baseline pruning methods. In addition, we demonstrate that our method preserves performance even in an unseen domain setting.
引用
收藏
页码:594 / 604
页数:11
相关论文
共 50 条
  • [41] Object detection network pruning with multi-task information fusion
    Li, Shengming
    Xue, Linsong
    Feng, Lin
    Wang, Yifan
    Wang, Dong
    WORLD WIDE WEB-INTERNET AND WEB INFORMATION SYSTEMS, 2022, 25 (04): : 1667 - 1683
  • [42] Offensive language identification with multi-task learning
    Zampieri, Marcos
    Ranasinghe, Tharindu
    Sarkar, Diptanu
    Ororbia, Alex
    JOURNAL OF INTELLIGENT INFORMATION SYSTEMS, 2023, 60 (03) : 613 - 630
  • [43] Multi-Task Learning for Multiple Language Translation
    Dong, Daxiang
    Wu, Hua
    He, Wei
    Yu, Dianhai
    Wang, Haifeng
    PROCEEDINGS OF THE 53RD ANNUAL MEETING OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS AND THE 7TH INTERNATIONAL JOINT CONFERENCE ON NATURAL LANGUAGE PROCESSING, VOL 1, 2015, : 1723 - 1732
  • [44] Offensive language identification with multi-task learning
    Marcos Zampieri
    Tharindu Ranasinghe
    Diptanu Sarkar
    Alex Ororbia
    Journal of Intelligent Information Systems, 2023, 60 : 613 - 630
  • [45] Residual based hierarchical feature compression for multi-task machine vision
    Chen, Chaoran
    Xu, Mai
    Li, Shengxi
    Liu, Tie
    Qiao, Minglang
    Lv, Zhuoyi
    2023 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA AND EXPO, ICME, 2023, : 1463 - 1468
  • [46] Natural Language Description of Video Streams Using Task-Specific Feature Encoding
    Dilawari, Aniqa
    Khan, Muhammad Usman Ghani
    Farooq, Ammarah
    Zahoor-Ur-Rehman
    Rho, Seungmin
    Mehmood, Irfan
    IEEE ACCESS, 2018, 6 : 16639 - 16645
  • [47] Task Relabelling for Multi-task Transfer using Successor Features
    Balla, Martin
    Perez-Liebana, Diego
    2022 IEEE CONFERENCE ON GAMES, COG, 2022, : 353 - 360
  • [48] Deductive competence: Mental models or task-specific strategies?
    Yule, P
    CAHIERS DE PSYCHOLOGIE COGNITIVE-CURRENT PSYCHOLOGY OF COGNITION, 1997, 16 (1-2): : 247 - 253
  • [49] On the Task-Specific Effects of Fragmentation in Modular Process Models
    Abbad-Andaloussi, Amine
    Schreiber, Clemens
    Weber, Barbara
    CONCEPTUAL MODELING, ER 2024, 2025, 15238 : 18 - 37
  • [50] Novice comprehension of programs using task-specific and non-task-specific constructs
    Engebretson, A
    Wiedenbeck, S
    IEEE 2002 SYMPOSIA ON HUMAN CENTRIC COMPUTING LANGUAGES AND ENVIRONMENTS, PROCEEDINGS, 2002, : 11 - 18