If you're getting 95% accuracy on training set, but only 75% on test set, this points to serious overfitting, which none of the measures you've listed are likely to address.
It's also suspicious that validation result are so close to training, but far from test. This often happens when you change validation set during training, meaning there's effectively no ...
I suggest to apply a nonrandom weight initialisation in order to see the impact of random initialization.
For instance, you can use the Nguyen-Widrow weight initialization.
Nguyen-Widrow initialization function
layer: core.Layer object
ci = layer.ci
cn = ...
You only need such a projection if you are using only dense layers for your model (i.e. a multilayer perception (MLP)). You can simply have a convolutional autoencoder, where the layers are convolutions and max pooling, and therefore the number of parameters is drastically reduced with respect to an MLP.
You can check Keras' tutorial on autoencoders, ...
It’s not wrong. NN training is inherently stochastic. As an optimisation problem, the tuning of a NN depends on the initialisation (initialisation of the weights). So the result (the local minimum you end up in) depends on the initialisation too.
There are mainly two ways to go :
if this is not a problem for your use case (if only the global performance ...
You can not feed your network with two inputs with different number of samples, and this also does not make sense.
You have 2 inputs with shape (502,) and (1002,) (You have said you want to extract features also from your second dataset). Let's consider the batch size is 1 for the sake of simplicity. So the model takes one sample each time to move it through ...