Blog
About


Data-Driven Struct Eng

Data-Driven Structural Engineering Research.
Unlocking Research Collabs.

 

Feel free to suggest new papers to the collection

2
recommends
+1 Recommend
4
shares

Editors

Editorials

  • Record: found
  • Abstract: found
  • Editorial: found
Is Open Access

Collabs with Abambres (check inside)

 Miguel Abambres (corresponding) (2018)
Would you like to co-author some papers with me? I´d love to! My role would be developing Artificial Neural Network (ANN) models and write that part of the paper, for any possible classification problem or nonlinear function you may wish to approximate. I will use an ANN-based software of my authory, developed during my postdoc in 2017/2018, which validation results are reported @ software validation report ( even though several improvements have been implemented since i finished the first version of the code ) For the above reason, the first papers (5 or 6) resulting from the application of that software to real problems are being submitted for publication until Christmas 2018. I will just need at least several hundreds of data points (inputs, target) from you, accurately obtained from either experiments or numerical methods, and the ANN will give us a matricial formula to accurately and very efficiently estimate the function of interest or classify examples - you can see this method as an advanced regression method formulated based on the brain behavior (the most powerful regression model in the world) The ANN application is to your problems, whatever you feel like….you just have to imagine situations from your present research interests where it would be useful to LEARN from examples, i.e. A ) Situations where computing something is quite COSTLY (time and/or resources) and UNFEASIBLE in practice (ie, when employing the traditional methods - experimental or numerical) and B) Situations where either (i) there´s no analytical model to predict values we want or (ii) the existing analytical models are too conservative or unsafe Whats the goal of my proposal concerning the application of Artificial Neural Networks to your problems? Imagine that you have a vectorial variable Y=(Y1,....,Ym) , where sub-functions Y1 to Ym are ALL qualitative OR ALL quantitative/numerical, determined experimentally or numerically ** as function of variables (X1, X2, X3, ..., Xn) that you are able to quantify or qualify (you can mix qualitative with quantitative Xi variables): Yj = Yj (X1, X2, X3, ..., Xn), j=1,...,m ** using a powerful software, which is accurate but has the drawback of being computationally expensive, i.e. takes more than a few seconds to obtain EACH result (each point of any nonlinear function) What my software is going to do, is to determine a matricial expression (function) that in a few seconds (often less than a second) is able to compute (X1,...Xn ; Y) with high accuracy for ANY given (X1,....Xn) that you provide as input. In order to obtain that mathematical model, you will have to give me at least hundred(s) of (X1,...,Xn; Y) data points accurately determined ( not much less than max(3^n,200) , where n is the number of input (independent) variables X1,...Xn). NOTE: These models work great for interpolation, so don´t send data and expect my models to give great predictions outside data domain - e.g., TIME SERIES PREDICTION is not handled by my software If you accept the collaboration, i´ll write all ANN-related part of the papers, and you can pick any scientific topic you wish - i will just need to know the data points (X1,...,Xn; Y1,....,Ym) and variable names (with the corresponding units - when existent and if n <= 20) - in due time i´ll send you a proper excel template for that purpose ( along with a .txt file explaining how to fill the .xlsx ) Software restrictions: n <= 200 m <= 20 Any questions, feel free to contact me anytime. Thank you so much :)
    Bookmark
    • Record: found
    • Abstract: found
    • Editorial: found
    Is Open Access

    Author Guidelines (check inside)

     Miguel Abambres (corresponding) (2018)
    Guidelines all articles in the collection must comply with: Short and appelaing titles In text ref format = author + year Length < 36 pp Small text density Results in the form of tables and graphs Small mean and max relative errors (%) when analythical methods are proposed Future guidelines (if I manage to have my own journal): MS word editor Single column text, 1.5 line spacing, 12 pt Times NR, level 1 headlines > 13 pt Only in text figures / tables no extra formatting required for submission Double-blind peer review Review time < 2 weeks Submission - Online Publication time lag < 6 months Open Access + 50 USD / publication ( not per co-author, per publication ! )
      Bookmark

      Collection Information