-1

I want to classify telecom devices: switches, routers, etc. I know that there are pre-trained model available online: https://github.com/tensorflow/models

  • Will it be possible to use transfer learning using those? Or do I have to use a pre-trained model of telecom devices?
  • Which one will you recommend me?
Aizzaac
  • 3,146
  • 8
  • 29
  • 61
  • It will probably be possible, yes, but it depends on many factors and you will have to carry out the research if you want to find out. There are plenty of tutorials, guides and other resources about this topic out there. Which base model is best depends on many things too, again there is information available about the strengths and weaknesses each may have. For SO, the question is too broad. – jdehesa Mar 12 '20 at 15:48
  • I am entry level in Deep learning. Can you point me to those resources about the topic? With all that information, for me it is an answer. – Aizzaac Mar 12 '20 at 15:51
  • 1
    A simple [Google search](https://www.google.com/search?q=tensorflow+transfer+learning) will take you to some good ones, and then there are books, moocs, blogs, YouTube channels, ... I haven't personally worked with visual data much, but you can start with [one of their tutorials](https://www.tensorflow.org/tutorials/images/transfer_learning) [on the topic](https://www.tensorflow.org/tutorials/images/transfer_learning_with_hub). – jdehesa Mar 12 '20 at 16:01

1 Answers1

1

1) Not necessarily, it will be helpful but we don't always use pre-trained weight to get better accuracy but a faster convergence time. So, definitely, it's always better to use a pre-trained weight if possible as the filters learnt by the model are somewhat useful feature extractors useful.

2) If you can find a pre-trained weight for telecom devices, there's no point to hesitate. That would give a faster convergence time + may boost the accuracy slightly.

Zabir Al Nazi
  • 10,298
  • 4
  • 33
  • 60