Science topic

Reservoir Computing - Science topic

Explore the latest questions and answers in Reservoir Computing, and find Reservoir Computing experts.
Questions related to Reservoir Computing
  • asked a question related to Reservoir Computing
Question
5 answers
I am using a reservoir computing architecture comprising of an echo state network as per the paper "Reservoir Computing Approaches for Representation and Classification of Multivariate Time Series" (https://arxiv.org/pdf/1803.07870.pdf)
Briefly, the architecture has four parts;
1. Reservoir module (echo state network)
2. Dimensional reduction module
3. Representation module
4. Readout module (linear regression, SVM, or MLP)
For a multivariate time series classification task that I am doing, keeping all parameters the same in parts 1-3 from above, when I use linear regression as readout, I get an F1 score of about 0.25 and AUROC of about 0.58. When I use MLP as readout, I get an F1 score of about 0.4 (+0.15 from linear regression) and AUROC of about 0.8 (+0.22 from linear regression).
Quoting from the paper "A Practical Guide to Applying Echo State Networks"(https://www.ai.rug.nl/minds/uploads/PracticalESN.pdf), section 3.1.;
"For classification tasks, input data u(n) which are not linearly separable in the original space RNu, often become so in the expanded space RNx of x(n), where they are separated by Wou"
My hypothesis for the difference in score between linear regression and MLP readout is that the output of the echo state network is not linearly separable and that's why MLP performs better as it is able to learn more complex patterns in the output. Is this correct?
If my hypothesis is correct, what could be done to make the echo state network output more linearly separable to improve the performance when using linear regression?
I would like to use linear readout as it has a shorter training time than MLP and from what I understood from the literature, it is more common to use linear readout when using echo state networks.
  • asked a question related to Reservoir Computing
Question
3 answers
Dear everyone,
Reservoir Computing(RC) is a newly-developed machine learning framework for time series prediction. However, cross validation is crucial for testing the overfitting problem. Unlike the multivariave linear regression model and ANN, the neuron state of RC has memory properity for the previous input data. So some traditional cross validation methods for data-driven prediction models might be not totally suitable for RC.
Would you mind providing me with some suggestions to make cross validation for RC?
Thank you!
Relevant answer
  • asked a question related to Reservoir Computing
Question
3 answers
Wolf's and Rosenstein's algorithms does not seem to include the multidimensional scenario (if I understand them correctely). I want to measure the nonlinear dynamics of the liquid state machine's dynamical system.
  • asked a question related to Reservoir Computing
Question
3 answers
Is there component correspondence between Reservoirs and Capsules? Do you think an architecture could implement both, either in different complementary modules or in a single integrated module? Could this architecture then be used for NLP tasks?
Maass W., Natschläger T. , and Markram H. (2002). “Real-Time Computing Without Stable States: A New Framework for Neural Computation Based on Perturbations"
Sabour S., Frosst N., Hinton, G.E., (2017). "Dynamic Routing Between Capsules"
Relevant answer
Answer
To clarify the relationships between ANNs and Reservoir Computing/Capsule Networks, you may have a look at:
  • asked a question related to Reservoir Computing
Question
3 answers
I am aware that the process itself depends on the geology of the reservoirs to a certain extent in addition to of course development scheme (depletion followed by water injection or gas injection .. etc). I am in the process of trying to compile a benchmarking of recovery vs. reservoir geology vs. mobility ratio vs. heterogeneity (areal and vertical) for different waterflood patters. Then I might try to develop further to other development schemes (gas injection, EOR .. etc)
I am sure that I need to use average drainage radius/permeability, communication between layers and BHP. However this is making it very complicated. 
Any ideas? Any suggested references? Any one who is interested in trying this as a first method of field development appraisal/bidding process?
What I am thinking is once these factors are estimated, a reasonable estimate of recovery factor can be achieved even before simulation process starts.  Since static and dynamic models are using basically this data to generate forecasts in the first place. Appreciate you feedback and discussion. 
Relevant answer
Answer
These incorporating parameters have significant effect on the recovery factor. Specially, Permeability that influences on the cut-off calculation