Do Kernel and Neural Embeddings Help in Training and Generalization?

被引:0
|
作者
Rahbar, Arman [1 ]
Jorge, Emilio [1 ]
Dubhashi, Devdatt [1 ]
Chehreghani, Morteza Haghir [1 ]
机构
[1] Chalmers Univ Technol, Dept Comp Sci & Engn, SE-41296 Gothenburg, Sweden
关键词
Kernel embedding; Gram matrix; Neural Network; Convergence;
D O I
10.1007/s11063-022-10958-8
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Recent results on optimization and generalization properties of neural networks showed that in a simple two-layer network, the alignment of the labels to the eigenvectors of the corresponding Gram matrix determines the convergence of the optimization during training. Such analyses also provide upper bounds on the generalization error. We experimentally investigate the implications of these results to deeper networks via embeddings. We regard the layers preceding the final hidden layer as producing different representations of the input data which are then fed to the two-layer model. We show that these representations improve both optimization and generalization. In particular, we investigate three kernel representations when fed to the final hidden layer: the Gaussian kernel and its approximation by random Fourier features, kernels designed to imitate representations produced by neural networks and finally an optimal kernel designed to align the data with target labels. The approximated representations induced by these kernels are fed to the neural network and the optimization and generalization properties of the final model are evaluated and compared.
引用
收藏
页码:1681 / 1695
页数:15
相关论文
共 50 条
  • [1] Do Kernel and Neural Embeddings Help in Training and Generalization?
    Arman Rahbar
    Emilio Jorge
    Devdatt Dubhashi
    Morteza Haghir Chehreghani
    Neural Processing Letters, 2023, 55 : 1681 - 1695
  • [2] Fast Neural Kernel Embeddings for General Activations
    Han, Insu
    Zandieh, Amir
    Lee, Jaehoon
    Novak, Roman
    Xiao, Lechao
    Karbasi, Amin
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 35, NEURIPS 2022, 2022,
  • [3] Neural Generalization of Multiple Kernel Learning
    Ghanizadeh, Ahmad Navid
    Ghiasi-Shirazi, Kamaledin
    Monsefi, Reza
    Qaraei, Mohammadreza
    NEURAL PROCESSING LETTERS, 2024, 56 (01)
  • [4] Neural Generalization of Multiple Kernel Learning
    Ahmad Navid Ghanizadeh
    Kamaledin Ghiasi-Shirazi
    Reza Monsefi
    Mohammadreza Qaraei
    Neural Processing Letters, 56
  • [5] Neural Tangent Kernel: Convergence and Generalization in Neural Networks
    Jacot, Arthur
    Gabriel, Franck
    Hongler, Clement
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 31 (NIPS 2018), 2018, 31
  • [6] Learning Kernel-Based Embeddings in Graph Neural Networks
    Navarin, Nicole
    Dinh Van Tran
    Sperduti, Alessandro
    ECAI 2020: 24TH EUROPEAN CONFERENCE ON ARTIFICIAL INTELLIGENCE, 2020, 325 : 1387 - 1394
  • [7] Neural Tangent Kernel: Convergence and Generalization in Neural Networks (Invited Paper)
    Jacot, Arthur
    Gabriel, Franck
    Hongler, Clement
    STOC '21: PROCEEDINGS OF THE 53RD ANNUAL ACM SIGACT SYMPOSIUM ON THEORY OF COMPUTING, 2021, : 6 - 6
  • [8] Redundant representations help generalization in wide neural networks
    Doimo, Diego
    Glielmo, Aldo
    Goldt, Sebastian
    Laio, Alessandro
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 35 (NEURIPS 2022), 2022,
  • [9] Redundant representations help generalization in wide neural networks
    Doimo, Diego
    Glielmo, Aldo
    Goldt, Sebastian
    Laio, Alessandro
    JOURNAL OF STATISTICAL MECHANICS-THEORY AND EXPERIMENT, 2023, 2023 (11):
  • [10] Trainable Calibration Measures For Neural Networks From Kernel Mean Embeddings
    Kumar, Aviral
    Sarawagi, Sunita
    Jain, Ujjwal
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 80, 2018, 80