The code in this directory reproduces the figures in:

Zhou Fan, Zhichao Wang, ``Spectra of the Conjugate Kernel and Neural Tangent Kernel for linear-width neural networks''

Figure numbers below correspond to the above manuscript.

- predict_spectrum.py:
    Implementation of algorithm that solves the fixed-point equations, with
    specializations to compute the limit spectra of the CK and NTK

- simulate_NN.py:
    Main script for simulating and analyzing the CK and NTK spectra at random
    initialization of the weights

- gaussian.py:
    Generates Figure 1 and i.i.d. Gaussian training data figures in Appendix J.4

- CIFAR10.py:
    Generates Figure 2 and CIFAR-10 figures in Appendices J.2 and J.4

- training/train.py and trained.py
    Generates Figure 3 and post-training figures in Appendix J.4

- orthogonality.py
    Generates figure in Appendix J.1

- CIFAR10_trained.py
    Generates figures in Appendix J.5

CIFAR-10 experiments use the data downloaded from:

https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz

