Finding useful inputs feature

Hi, I’m new to Pytorch. I have a learned model to predict obesity. I wish to know that which feature are really useful for the network.

Here is my model.

class Net(nn.Module):
        def __init__(self):
            super(Net, self).__init__()
            self.fc1 = nn.Linear(559, 100)
            self.fc2 = nn.Linear(100, 2)
            self.softmax = nn.Softmax(dim=2)
        def forward(self, x):
            x = F.relu(self.fc1(x))
            return self.softmax(self.fc2(x))
net = Net().double()
net = net.cuda()

Is there anyway I can find which feature is useful?

I look for many solution and one of it said that I can sum up the gradient during the backpropagation step, but how and why?