Proceedings of the 13th EAI International Conference on Mobile Multimedia Communications, Mobimedia 2020, 27-28 August 2020, Cyberspace

Research Article

Deep N-ary Error Correcting Output Codes

Download374 downloads
  • @INPROCEEDINGS{10.4108/eai.27-8-2020.2299197,
        author={Hao  Zhang and Joey Tianyi Zhou and Tianying  Wang and Ivor W. Tsang and Rick Siow Mong  Goh},
        title={Deep N-ary Error Correcting Output Codes},
        proceedings={Proceedings of the 13th EAI International Conference on Mobile Multimedia Communications, Mobimedia 2020, 27-28 August 2020, Cyberspace},
        publisher={EAI},
        proceedings_a={MOBIMEDIA},
        year={2020},
        month={11},
        keywords={deep n-ary ecoc ensemble learning multi-class classification},
        doi={10.4108/eai.27-8-2020.2299197}
    }
    
  • Hao Zhang
    Joey Tianyi Zhou
    Tianying Wang
    Ivor W. Tsang
    Rick Siow Mong Goh
    Year: 2020
    Deep N-ary Error Correcting Output Codes
    MOBIMEDIA
    EAI
    DOI: 10.4108/eai.27-8-2020.2299197
Hao Zhang1,*, Joey Tianyi Zhou1, Tianying Wang1, Ivor W. Tsang2, Rick Siow Mong Goh1
  • 1: Institute of High Performance Computing, A*STAR
  • 2: AAII, University of Technology Sydney
*Contact email: zhang_hao@ihpc.a-star.edu.sg

Abstract

Ensemble learning consistently improves the performance of multi-class classification through aggregating a series of base classifiers. To this end, data-independent ensemble methods like Error Correcting Output Codes (ECOC) attract a increasing attention due to its easiness of implementation and parallelization. Specifically, traditional ECOCs and its general extension N-ary ECOC decomposes the original multi-class classification problem into a series of independent simpler classification subproblems. Unfortunately, integrating ECOCs, especially N-ary ECOC with deep neural networks, termed as deep N-ary ECOC, is not straightforward and yet fully exploited in the literature, due to high expense of training base learners. To facilitate training of N-ary ECOC with deep learning base learners, we further propose three different variants of parameter sharing architectures for deep N-ary ECOC. To verify the generalization ability of deep N-ary ECOC, we conduct experiments by varying the backbone with different deep neural networks architectures for both image and text classification task. Furthermore, extensive ablation studies on deep N-ary ECOC show its superior performance over other deep data-independent ensemble methods.