I would appreciate if you could please advise as to where I can find details on design guidelines for the CNN components (Convolution, Pooling, Transe Convolution, the number of filters to use…etc).
It all seems a black box so far. I understand that the filters are learnt but what features are they extracting and how. How to decide how many of them and in what order for a particular task?
All the literature seems to be pointing to is trial and error in terms of the sequence of blocks and the number of those blocks to be used along with a binary number (2^) of filters which tend to double in every stage.
Grateful for more information on the topic.
It is mostly a trial and error process that’s why. You can use similar model to your application and searching techniques to maximize performance but still its a trial and error process.
Noted thanks for your quick reply.
A quick one, what do you guys recommend taking next after finishing DLS:
Tensorflow Developer Prof. Certifcate
After DLS better take TF developer dhe MLOps needs mos advanced knowledge.
As a follow up to this thread, I skimmed through the first 3 weeks of Tensorflow Developer Prof. Certifcate, and it seems that the majority of the ideas and implementation in this course are already covered in DLS.
I am writing this to seek input/guidance whether there is still value in Tensorflow Developer Prof. Certifcate set of courses or should I go straight to NLP Specialization.
There are additional concepts in the TF stream of courses for sure and tensorflow is very widely used in machine learning these days.