Raul Hector Cajias

Questions and Answers (5) View all

• Answer added in Artificial Intelligence
  44 Does anyone know some procedure that it is impossible to be carried out by a Turing machine?
  By Juan-Esteban Palomar Tarancon · I am retired
  Raul Cajias · ResearchGate

  If I understand correctly, you're asking for examples of undecidable problems? The halting problem is good example: Can a routine be written to deter... [more]

  If I understand correctly, you're asking for examples of undecidable problems? The halting problem is good example: Can a routine be written to determine wether a second routine will ever finish, given some input? This has been proven to be undecidable
  Following
• Answer added in Complex Systems
  19 Which program do you use for the generation layouts for networks of thousands of nodes?
  By Francesco Rao · Universität Freiburg
  Raul Cajias · ResearchGate

  You might want to look into GraphViz. If I remember correctly, you describe the graph using their special notation which they can then render. If yo... [more]

  You might want to look into GraphViz. If I remember correctly, you describe the graph using their special notation which they can then render. If you can export the graph to graphml format there are some R packages than could be useful as well.
  Following
• Answer added in Computer Science
  26 After a neural network is trained. Can it really be validated?
  By Daniel Crespin · Central University of Venezuela
  Raul Cajias · ResearchGate

  Hello Daniel, I'm still unclear how is it that there is no overtraining if test data is classified with 100% accuracy. The fact that A' and B' are cl... [more]

  Hello Daniel, I'm still unclear how is it that there is no overtraining if test data is classified with 100% accuracy. The fact that A' and B' are classified with 0% accuracy makes this problem more apparent. Have you tried to varying degrees of accuracy? and out of curiosity, is this supervised or unsupervised learning? I'm on mac so I can't run your exe.. however I did not find any training data (csv, txt). Did I miss it? what's the file name? On the other hand, just to go back to the initial question, I think we can all agree neural networks (and any ML technique) can and should be validated with test data.. What you score on the test data is the final score. Not the training. If you haven't yet, I recommend UC Irvines extensive dataset repository. It's a great benchmark for ML algorithms : http://archive.ics.uci.edu/ml/datasets.html Cheers
  Following
• Answer added in Computer Science
  26 After a neural network is trained. Can it really be validated?
  By Daniel Crespin · Central University of Venezuela
  Raul Cajias · ResearchGate

  Hi, I'm not sure if I understood your question, but I can tell you that usually a match of 100% on the training data means you're overfitting. Overfi... [more]

  Hi, I'm not sure if I understood your question, but I can tell you that usually a match of 100% on the training data means you're overfitting. Overfitting means you have fine-tuned your NN to the point where slight changes in the data will commpletly throw it off. To properly train your NN you may want to look at bagging/boosting, and find a confortable margin of error in the success of you NN. Is 80% good enough? 70%? The wider the error margin, the more A', A'' can vary from the original A. As far as testing, you cant rely on the performance of the training data to tell you how good your NN is (see overfitting). Instead it has to be an iterative process of selecting good training sets (bagging/boosting) training your NN (to a confortable error margin) and seeing how it performs on new data. I hope this helps, and I apologize in advance if I didn't understand your question and just told you stuff you already knew ;) Cheers
  Following
• Answer added in Social Network Analysis
  2 Twitter Research
  By Omer Ajmal · The Islamia University of Bahawalpur
  Raul Cajias · ResearchGate

  A lot of the stuff I've seen has to do with influence and meme propagation. Sentiment analysis is also quite popular. If you're looking for a place t... [more]

  A lot of the stuff I've seen has to do with influence and meme propagation. Sentiment analysis is also quite popular. If you're looking for a place to start, the book "Mining Twitter" published by Oreilly has a good outline of tools/approaches commonly used.
  Following