could someone please give me a logical reason why are we doing these 3 methods like whats the logic behind doing these three models, andrew explains how to do it not why to do it.
Hi, @Saim_Rehman Thank you for your question.
There are several advantages to using these models.
For instance, the Inception model addresses the challenge of building deeper and more computationally efficient neural networks. Through its filters from different sizes, it manages to create more realistic representations without increasing the number of parameters and the computational complexity.
If interested I invite you to read the paper where their main advantages are addressed:
“ImageNet Large Scale Visual Recognition Challenge” ([1409.0575] ImageNet Large Scale Visual Recognition Challenge, it has almost 40,000 citations 
)
We use mobileNets because for some applications we do not have the hardware to run powerful models on-the-fly. Therefore they are efficient on mobile devices or embedded devices like Raspberry Pi, Espressif, etc.
In the case of ResNet, it addresses the vanishing gradient problem that occurs in deeper NN. Therefore you can train faster and more efficient these models.
yea i know what it is used for, but i just want to know why are these things working, like using 1x1 convolutions and things like these why are these working
Hi, @Saim_Rehman you asked why in the first question, I am sorry If I got wrong what you mean to ask.
The 1x1 convolutions enable the Inception module to have flexibility in merging features from different filter sizes. They are used to combine all the information across different scales from the filters sizes.