nasadem | NASADEM SRTM and TanDEM-X products | Runtime Evironment library

The histograms you see are weight distributions of two dense layers.
Every dense layer has weights(w1, w2, .., wn) and a bias(b). The weights are called kernels in Tensorflow.

The histograms represent kernels and biases of two Dense layers.

How to interpret them?

As this is 3d histogram first it contains three axis.

x-axis - The actual value the kernels and biases take. ie value of b,w1,w2 etc
y-axis - The epoch number.
z-axis - The number of weights

We will take dense_1/bias_0 for our example. In this
x-axis : -0.8 to 1.6
Interpretation - All the values in the bias vector of the dense_1 layer are between -0.8 and 1.6.

y-axis : 0 - 22
Interpretation - The histogram shows values for each epoch from 0 to 22.

z-axis :
Interpretation - It shows how many biases have the value represented by x-axis. ie. If more biases have a value equal to zero you will get the highest peak (highest value of z) at x = 0.

So from the histogram for dense_1/bias_0 you can interpret that.
At epoch = 0 , all the biases had value 0. Since you get a peak at x = 0 and no z value for other values of x.

At epoch = 22, your bias vector does not contain any value between 0.0 and 0.4.

Why are these histograms used?
These histograms are used to see whether the model is actually learning or not. If the histogram remains the same for every epoch, then your model is not training properly and there is some error.

Useful links:

1. Tensorboard Tutorial
2. Using Tensorboard for debugging