Corpus ID: 219721265

Kernel Alignment Risk Estimator: Risk Prediction from Training Data

@article{Jacot2020KernelAR,
  title={Kernel Alignment Risk Estimator: Risk Prediction from Training Data},
  author={Arthur Jacot and Berfin cSimcsek and Francesco Spadaro and Cl{\'e}ment Hongler and Franck Gabriel},
  journal={ArXiv},
  year={2020},
  volume={abs/2006.09796}
}
  • Arthur Jacot, Berfin cSimcsek, +2 authors Franck Gabriel
  • Published 2020
  • Mathematics, Computer Science
  • ArXiv
  • We study the risk (i.e. generalization error) of Kernel Ridge Regression (KRR) for a kernel K with ridge λ > 0 and i.i.d. observations. For this, we introduce two objects: the Signal Capture Threshold (SCT) and the Kernel Alignment Risk Estimator (KARE). The SCT θK,λ is a function of the data distribution: it can be used to identify the components of the data that the KRR predictor captures, and to approximate the (expected) KRR risk. This then leads to a KRR risk approximation by the KARE ρK… CONTINUE READING

    References

    Publications referenced by this paper.
    SHOWING 1-10 OF 32 REFERENCES

    A Random Matrix Approach to Neural Networks

    VIEW 4 EXCERPTS
    HIGHLY INFLUENTIAL

    Optimal Rates for the Regularized Least-Squares Algorithm

    VIEW 4 EXCERPTS
    HIGHLY INFLUENTIAL

    Ridge Regression: Structure, Cross-Validation, and Sketching

    VIEW 4 EXCERPTS
    HIGHLY INFLUENTIAL

    Effective Dimension and Generalization of Kernel Learning

    VIEW 2 EXCERPTS

    Enhanced Convolutional Neural Tangent Kernels

    VIEW 1 EXCERPT