Complexity of Representations in Deep Learning

被引:1
|
作者
Ho, Tin Kam [1 ]
机构
[1] IBM Watson Hlth, Yorktown Hts, NY USA
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 50 条
  • [1] Learning Deep Parsimonious Representations
    Liao, Renjie
    Schwing, Alexander
    Zemel, Richard S.
    Urtasun, Raquel
    [J]. ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 29 (NIPS 2016), 2016, 29
  • [2] Learning Input Features Representations in Deep Learning
    Mosca, Alan
    Magoulas, George D.
    [J]. ADVANCES IN COMPUTATIONAL INTELLIGENCE SYSTEMS, 2017, 513 : 433 - 445
  • [3] Learning Deep Representations for Photo Retouching
    Li, Di
    Rahardja, Susanto
    [J]. IEEE TRANSACTIONS ON MULTIMEDIA, 2024, 26 : 3153 - 3163
  • [4] Geometric deep learning on molecular representations
    Atz, Kenneth
    Grisoni, Francesca
    Schneider, Gisbert
    [J]. NATURE MACHINE INTELLIGENCE, 2021, 3 (12) : 1023 - 1032
  • [5] Geometric deep learning on molecular representations
    Kenneth Atz
    Francesca Grisoni
    Gisbert Schneider
    [J]. Nature Machine Intelligence, 2021, 3 : 1023 - 1032
  • [6] Learning Deep Representations for Graph Clustering
    Tian, Fei
    Gao, Bin
    Cui, Qing
    Chen, Enhong
    Liu, Tie-Yan
    [J]. PROCEEDINGS OF THE TWENTY-EIGHTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, 2014, : 1293 - 1299
  • [7] Deep Learning with Relational Logic Representations
    Sourek, Gustav
    [J]. PROCEEDINGS OF THE TWENTY-EIGHTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE, 2019, : 6462 - 6463
  • [8] Deep Learning of Orthographic Representations in Baboons
    Hannagan, Thomas
    Ziegler, Johannes C.
    Dufau, Stephane
    Fagot, Joel
    Grainger, Jonathan
    [J]. PLOS ONE, 2014, 9 (01):
  • [9] Deep Learning to Hash with Multiple Representations
    Kang, Yoonseop
    Kim, Saehoon
    Choi, Seungjin
    [J]. 12TH IEEE INTERNATIONAL CONFERENCE ON DATA MINING (ICDM 2012), 2012, : 930 - 935
  • [10] Learning deep representations via extreme learning machines
    Yu, Wenchao
    Zhuang, Fuzhen
    He, Qing
    Shi, Zhongzhi
    [J]. NEUROCOMPUTING, 2015, 149 : 308 - 315