Whether ANN (artificial neural network) is more reliable than CCRD based optimization (Design Expert)?I have come across some papers and also during discussion in some conference, got some questions for which I am trying to find a precise solution.
Some used CCRD first for optimization of process parameters to maximize the product yield, then fed experimental result to ANN, fine tuned optimal condition.
Is that a right method?
How much reliable is ANN for optimization of process parameters?
Is it possible to use mere ANN for optimization (with actual experiments like CCRD to feed data)?
(P.N.: I have least knowledge in ANN)
Dear Venkatesh Kamath,
Thank you for your interesting question. I think that the ANN is one of the most powerful tool not only in the scientific field but also in the engineering field.
If we focus on the supervised ANNs, as you know, the ANN is a net which combines groups of nonlinear functions. Depending on the transfer functions used (sigmoid, hyperbolic, and so on,) one can estimate any dependent variable using the adequate independent variables in a system given.
In this case, depending on the complexity of the system, the user can adapt the ANN topology in order to adapt it to the complexity of the system. In this way, one can achieve lower estimation errors the output variable.
Once the ANN has been optimized and successfully validated, one can estimate the independent variable in the range of training values initially established. In this moment you can easily optimize the system by a simple nonlinear interpolation using the model optimized.
Following similar procedures, the supervised ANN's are undoubtedly one of the most powerful tool in the optimization fields, among other important applications. Although I have attached a publication that is being important in the nanosensor field, apart from this, In the last years, we have published a high number of papers in this field, please take a look to my publications and If you have any questions, feel free to contact me.
Thank you once again for your question.
Cheers,
José
I work on NP Hard problems. Needless to say that the problems come under the category of optimization problems. I have found that ANNs are good for regression where as GAs and DGAs work better in case of optimization. Though Jose is much more experienced than me, however, I have stated what I found in my experiments. Moreover, the choice depends on the type of problem you are working on.
Thank you Prof. Jose S. Torrecilla, Prof. Harsh Bhasin & Prof. Abdul Kayom Md Khairuzzaman.
I had done net search on this, plus from your feedback, I am convinced that ANN is superior and undoubtedly very strong in analysing Non linear multivariate systems.
Other than MATLAB is there any tool (free downloadable / free online) available for performing such analysis?
Any suggestions welcome.
Dear Venkatesh Kamath,
Please think on designing your own ANN by yourself. Writing your own ANN code is not so complicated. Regarding your question, there are a big free software offer, please take a look to this link.
I hope this website will be helpful to you.
Cheers,
José
http://www.winsite.com/neural/neural+networks+download/
- Badges
- Science topic