CODEWITHSUNDEEP
pythonpytorchneural-networkdistribution
intern5
intern5

Reputation: 19

Pytorch neural net transforming 1d uniform distribution to 2d gaussian

I need to train a neural network to convert a 1d uniform distribution to a 2d gaussian distribution.

I've constructed a simple feedforward network with leaky ReLu activation and trained using MMD as the loss function

d1_to_d2 = nn.Sequential(
    nn.Linear(1, 64),
    nn.LeakyReLU(),
    nn.Linear(64,64),
    nn.LeakyReLU(),
    nn.Linear(64,64),
    nn.LeakyReLU(),
    nn.Linear(64, 2),
)

Where it takes a 1-d point and outputs a 2d one. However the output distribution keeps changing to a

NN output distribution

Changing the depth of the network, number of weights, minibatch size, activation functions, and regulisation dont seem to change anything and I keep getting this general shape. Is there something obvious that I am missing?

Upvotes: 0

Views: 43

Answers (0)

Related Questions