1

Topic: Adaptation of model to the new data.

Rather popular question in   about  - "and how to modify a network so that she would recognize the data from other allocation, on which it not .". Met interesting (theoretical) operation on this subject which would like to share. Domain adaptation

2

Re: Adaptation of model to the new data.

Hello, SomeOne_TT, you wrote: SO _> rather popular question in   about  - "and how to modify a network so, SO _> that she would recognize the data from other allocation, on which it not .". And Transfer learning not this problem solves?

3

Re: Adaptation of model to the new data.

Hello, Nuzhny, you wrote: N> Hello, SomeOne_TT, you wrote: SO _>> rather popular question in   about  - "and how to modify a network so, SO _>> that she would recognize the data from other allocation, on which it not ." . N> And Transfer learning not this problem solves? No. Transfer Learning is a process of change of the trained model in one area of knowledge for usage in another. In this process truncating of the last layers of a network is quite normal. For example, usage of a deep network  persons for recognition of cars - without cropping not to manage. Domain Adaptation is usage of the trained network on  with other properties (allocation). We tell, if there is a network trained on jokes Domain Adaptation it is necessary for its usage on news or . Words both there and there like identical, but without adaptation results will be unimportant. P.S. I will write once. I not the welder, I only study and I can be mistaken.