Deep Learning networks (such as SdA) have been shown very suitable for many Pattern Recognition applications, however it is said that Autoencoders can't generalize well over noisy data like in typical Financial Time series.
Would be there any other Deep Learing technique that has been shown to have a good performance under a noisy dataset ?
Mr Parsapoor
Thank you very much for the reference. Would you have a link for a toolbox/library repository for the referred code ? (regardless of the used programming language).
Best Regards.
Hello Jorcy,
Regarding time series and temporal data, I think a good approach is to use a recurrent neural network. Networks that use Long Short-Term Memory cells ( http://people.idsia.ch/~juergen/rnn.html ) were shown to have very good results.
There is some work on deep recurrent denoising autoencoders which might be of interest (e.g. see here: http://www1.icsi.berkeley.edu/~vinyals/Files/rnn_denoise_2012.pdf)
It is also worth mentioning the Hierarchical Temporal Memory architecture by Numenta which seems to deal with nonstationary data. Here is the whitepaper: http://numenta.com/assets/pdf/whitepapers/hierarchical-temporal-memory-cortical-learning-algorithm-0.2.1-en.pdf
Is is possible to denoise your problem using other methods? While time series analysis is not my field, I heard that there are methods that take sequential snapshots and consecutive differences in order to convert a nonstationary problem to a stationary one. I think this might be related to "denoising" the time series data.
If you are interested in univariate time series data predictions, have a look in PSF algorithm and it's R Packages available at :
CRAN: https://cran.r-project.org/web/packages/PSF/index.html
GitHub: https://github.com/neerajdhanraj/PSF
How to use: https://www.researchgate.net/publication/304131481_PSF_Introduction_to_R_Package_for_Pattern_Sequence_Based_Forecasting_Algorithm
and
https://www.researchgate.net/publication/304580701_Introduction_of_seasonality_concept_in_PSF_algorithm_to_improve_univariate_time_series_predictions
For further details contact me at: http://www.neerajbokde.com
Article PSF: Introduction to R Package for Pattern Sequence Based Fo...
Article Introduction of seasonality concept in PSF algorithm to impr...
Hi,
I won't suggest recurrent neural networks as the leading approach since presently they are hard to train and easily overfitted, and most of their functionality can be performed by other models. The conditional restricted Boltzmann machine is what you're looking for. Look at the thesis titled "Nonlinear multilayered sequence models" by Ilya . I'm writing a survey on time series using deep learning, should be ready soon.
- Badges
- Science topic