Matlab tools for Deeep Learning allow us to solve classification tasks with static features (e.g., pixels values of images) and tasks with dynamic features (time-series). For time-series it is recommended to use LSTM (Long Short Time Memory) layer. But how to build model, if we want to use simultaneously time-series parameters and some static parameters?
Naturally, it is possible to create a fictitious time series (with constant values) from each value of a single static parameter, but this will be very inefficient.
Thanks for your answers beforehand. Regards, Sergey.
Sergey Porotsky dynamic features are first and second derivatives of the static features.
Dear Yosra Mohammed, you are right, but how it helps for my problem ?
Depends on the task but, suppose you want to classify an event that has dynamic and static features, you can treat dynamic features with RNN and static features with a Feed Forward Neural Network.
Than you can combine the two obtained "internal representations" (output of the LSTM and of the Neural Network) and connect it to a final FFNN that performs the classification.
E.g. look at https://arxiv.org/pdf/1911.05702.pdf - fig.1
Hi Sergey,
Have you taken a look at so-called spatial-temporal deep learning models?
A survey about such models is avalable by following:
- Wang et al., " Deep Learning for Spatio-Temporal Data Mining: A Survey ", 2019 - https://arxiv.org/pdf/1906.04928.pdf
It is quite possible to incorporate higher dimensional topological features into these models. To get a better idea, refer to the following papers:
- Gonzalez-Diaz et al., " Topological Tracking of Connected Components in Image Sequences ", 2018 - https://arxiv.org/pdf/1801.00939v1.pdf
- J.R. Stroud, " Dynamic Models for Spatio-Temporal Data ", 2001 - http://jrs390.georgetown.domains/papers/spacetime.pdf
- Fernández Abrevaya et al., " A 3D+t Laplace operator for temporal mesh sequences ", 2016 - https://hal.inria.fr/hal-01318763/document
Regards
Dear Amedeo Buonanno, thanks a lot for youe answer. I will carefulyy study your link.
Comments by the above experts and you are very fruitful, I will just refer the links for dynamic model based deep learning examples here : https://people.eecs.berkeley.edu/~ronf/PAPERS/anagabandi-icra18.pdf
Article CNNTracker: Online discriminative object tracking via deep c...
https://web.stanford.edu/~yplu/DynamicOCNN.pdf
Static models : Article Deep EHR: A Survey of Recent Advances on Deep Learning Techn...
https://github.com/tthtlc/awesome-source-analysis
- Badges
- Science topic