Semiautonomous seminar at University of Berkeley

I gave a talk on "convergence of overparametrized networks"(Min et al., 2021; Min et al., 2023) at the semiautonomous seminar, University of Berkeley. (host: Shankar Sastry)

Reference

  1. lin_conv_two_layer.png
    Convergence and Implicit Bias of Gradient Flow on Overparametrized Linear Networks
    Hancheng Min, Salma Tarmoun, René Vidal, and Enrique Mallada
    Journal of Machine Learning Research, 2023
    in preparation
  2. lin_conv_two_layer.png
    On the Explicit Role of Initialization on the Convergence and Implicit Bias of Overparametrized Linear Networks
    Hancheng Min, Salma Tarmoun, René Vidal, and Enrique Mallada
    In The 38th International Conference on Machine Learning (ICML), Jul 2021