I have time series signals that are VERY high dimensional (14000 data points), the number of samples is around 1000 samples. I want to use the Convolution neural networks to classify the time series signals into two classes. The input to the cnn is (1000*14000). Is it possible to perform such a classification task especially if the dimension of the input is much much higher than the number of samples that I have?? PLEASE ADVICE
$\begingroup$
$\endgroup$
-
$\begingroup$ AS @liangjy poitned out, you should experiment with RNNs as they are a more natural model for time series data. Additonally, a RNN model can be much smaller as the inputs are sequential and not parallel. $\endgroup$ – Thomas W May 4 '17 at 8:28
-
$\begingroup$ Question sounds perfectly clear to me, so I'm voting to leave open. $\endgroup$ – Firebug May 4 '17 at 11:07
$\begingroup$
$\endgroup$
You would need some kind of regularization technique (weight decay, dropout) in order to make sure that you're not overfitting your training data. In addition, I would suggest using recurrent neural networks instead of convolutional neural networks, since RNNs are a more natural model for time series data.
