0
$\begingroup$

I am trying to use a neural network to predict the next state output given the current state and action pairs. Both input and outputs are continuous variables. Due to the high dimensionality of each input, ( ~50 dimensional input ) and 48 dimensional output, I am not able to achieve an achieve a satisfiable enough accuracy.

I am thinking of using an auto-encoder to learn a latent representation of the state. Would a latent representation from an auto-encoder help to improve the prediction accuracy ? and can the latent representation have a higher dimensional space compared to the original state ?

$\endgroup$
1
$\begingroup$

i feel , i can answer this question, based the web source that i found and i read. that can use an autoencoder with hight latent representational space. example use LSTM autoencoder, LTSM autoencoder used for sequence or time series data. on the source, they use CNN autoencoder to denoise some syntetic noised data they have generated. but they have asked what is the meaning of this latent representation space? for what they have done before. on that answer. they said"your input data is noisy sinewave data. your are not supposed to use Convolutional autoencoder for sequence data".

https://stackoverflow.com/questions/59438488/cnn-autoencoder-latent-space-representation-meaning

addition other source for answer this question https://towardsdatascience.com/understanding-latent-space-in-machine-learning-de5a7c687d8d https://towardsdatascience.com/deep-inside-autoencoders-7e41f319999f

hopefully help answer :))

| improve this answer | |
$\endgroup$
  • $\begingroup$ hey thanks for your answer. I found out what I was looking for. I was looking more at a sparse autoencoder, where the activations in the hidden layer ( > input dimens ) is penalised to prevent the overfitting of the identity function $\endgroup$ – calveeen Apr 22 at 2:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.