Complexity of Representations in Deep Learning

被引:2
作者
Ho, Tin Kam [1 ]
机构
[1] IBM Watson Hlth, Yorktown Hts, NY USA
来源
2022 26TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION (ICPR) | 2022年
关键词
Keywords-representation learning; deep learning; classification; data complexity;
D O I
10.1109/ICPR56361.2022.9956594
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Deep neural networks use multiple layers of functions to map an object represented by an input vector progressively to different representations, and with sufficient training, eventually to a single score for each class that is the output of the final decision function. Ideally, in this output space, the objects of different classes achieve maximum separation. Motivated by the need to better understand the inner working of a deep neural network, we analyze the effectiveness of the learned representations in separating the classes from a data complexity perspective. Using a simple complexity measure, a popular benchmarking task, and a well-known architecture design, we show how the data complexity evolves through the network, how it changes during training, and how it is impacted by the network design and the availability of training samples. We discuss the implications of the observations and the potentials for further studies.
引用
收藏
页码:2657 / 2663
页数:7
相关论文
共 16 条
  • [1] [Anonymous], 2018, P 32 INT C NEUR INF
  • [2] Spectral Metric for Dataset Complexity Assessment
    Branchaud-Charron, Frederic
    Achkar, Andrew
    Jodoin, Pierre-Marc
    [J]. 2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2019), 2019, : 3210 - 3219
  • [3] Garcia LPF, 2018, INT C PATT RECOG, P874, DOI 10.1109/ICPR.2018.8545110
  • [4] github, About us
  • [5] Deep Residual Learning for Image Recognition
    He, Kaiming
    Zhang, Xiangyu
    Ren, Shaoqing
    Sun, Jian
    [J]. 2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2016, : 770 - 778
  • [6] Ho TK, 2002, IEEE T PATTERN ANAL, V24, P289, DOI 10.1109/34.990132
  • [7] Ioffe S., 2015, P 32 INT C MACHINE L, P448
  • [8] Jacot A, 2018, 32 C NEURAL INFORM P
  • [9] Decision Boundary of Deep Neural Networks: Challenges and Opportunities
    Karimi, Hamid
    Tang, Jiliang
    [J]. PROCEEDINGS OF THE 13TH INTERNATIONAL CONFERENCE ON WEB SEARCH AND DATA MINING (WSDM '20), 2020, : 919 - 920
  • [10] Krizhevsky A., 2009, Technical report, P1