The ingredients are going to a pretty high because Lord off changes will happen which is very near without putting So this is what is your listening next Uh that’s on You mean one of the less, Yeah So this is orders Important points So the architectural off arrested 30, for now, You have many worsens off This This is arrested 34 that is resting at 1 50 has invested 50 years The concept has changed Only thing is the number of layers that you are going to see here with British the sleepless Okay all our tree Crossley convolutions and same convolutions No change So we keep uniformity Ah keep it simple and design off the clock because we don’t want to play around.
I want my output to work on any of these convolutions I keep it as simple as possible Okay on if you observed that I know fully connected lists and I think that said that’s all on One thing to remember is the dotted lines in this case when the damage Okay one weathers my daughter main Okay so there there there are two things I’m not sure if you can see the dotted lines If your input and output are kind off racing each other that means you can switch or some of these lines that these are optional connections that the abuse it’s not necessary to take each one of them interest So if you are designing it that you are surpassing only one your little you can have some of them as dropouts saying that we don’t need them.
This other daughter So we will see I will implement one of these restaurants It’s not in our course It’s not in our standard for Macchia but I don’t believe this I would implement this rest in it and I will show you once you’re comfortable with the 1st 3 Algarve Networks Victory Lewis to Let’s see how we can design status So the first thing is well designed Second thing is will directly important bits and use them Perfect So this is orders the expanded our zoom diversion off arrested 34 Yeah 34 presents a total number of deep How deep the networks No coming to how to decide how many layers to bacon on the spot.
If it’s so very simple If you have around five years in a dog it’s a very easy network Okay so it’s not a leaving a drop it all If you have a known painless including initializing er shelling batch normalization isn’t good I can see you are still an easy one but a little complex Little leave us comfort to fight If you go to 30 and 100 these are the real evening looks especially answer that this one is very leave And in this network if you are using this your guide and if I have to skip the connections because the time is going to be very long So if we’re planning to have 100 layers are you in for delays Please make sure that you have to skip the connection The one which you didn’t dress so more of The less I can see these are examples of the rest nerds LSD ends on all this fun on anything about 1000 years I’m not sure I have never seen in my life.
We have no idea how deep is going to be So if you want Oh look at one of the layers which are very deep This could be an example of the inception It from Mobile Hey look at this And they have a lot of connection Lot of skipping is happening here So what is it The inception Adore consists off uncoordinated blocks off inception models The name inception was taken from the mainly made You need us sometimes A Maxwell Bloch is used even before the inception morning to reduce that dimension off So please remember it is not always necessary to take the image as it is the way we used to do BC the same thing began word heroes now ever shown you many times how to reduce the dimension in image processing No no Have I shown you anything on our toe in quarters Anything No night No right shortly.
If you remember is true It’s really PC An essay and a TSA I’m not sure you when you have done that Um have you heard of 40 years Any days I’m sure you one more dimensional election technique in machine learning today But apart from that if you want to do the same thing in your little how should we do it The first day was nothing That Your convolution factor solution was something like um sinking the day tights But if you look at convolution as the direction there is a problem let’s see how to notice it is the image you are having here which is the same 64 across 64 and after convolution,, before flattening you got any made savages Five Cross life, Arguelles, I would not say image I got some data which is five cross faith What we have done The way we can shrink this up.
Now can I use this for future processing If I give it to one-off our networks Saline and Alexa Donald will Alexe alienate be able to identify that this is a part of a lot of that image like this Yes I know Can I do something like this which I call the dimensional reduction Is it possible Think about yes we can do that Okay we got one Yes Anybody else thinks it’s not a good idea is a good idea, Okay It’s possible This Okay fine From my side You guys can try this out if you can do it If you put a CNN at the end Yes it is Possibility it will become your CNN again if you put enough CNN here.
What if I want to send this to some other networks Other networks will not have the recordings yet You remember We have done some including here You have used some freighters toe in court This data every given them the d quarterback It is like that cipher I’m not sure if you heard about this Let us say you want to secretly communicate something That you do is you implement your cipher There will be a filter with you Multiply with all the things and could create some new numbers Look like junk data Now this Seifert record this you have to give Divide the whole later by the cipher again Then you are going to get it back Take No this is what I call it is a cipher Whatever we have done here are we providing safer along with that So tomorrow if you want to recreate this image use the safe.
We’re not doing it right So what if I want to reduce the size of the inmates and if I want to recreate it we should be even able to replicate it back OK But in this case, what I will lose if this was 64 close 64 I cannot really replicate the whole thing back Cannot have the whole sample England So what I will do it,, Yes If I want to reduce the size to 50% I eternally did custody But still I will be let us say if the number seven is where I will be able to Wrigley Reapers and seven back This is called orto in goodness.
I’ll show you pretty one implementation also but a very high level This is how it looks like And if you ask me what is present inside these ah networks there are two things which are present one It’s called encoding Okay in court And when is equaled Anybody here from electron ICS and communication background or communication background or at least special DST part of anybody they should Theo cameras on Yeah anybody something’s background even from computer science background You guys would have done it,, Uh and Korda’s recorders No cryptology or something I have Yeah you can court starting Correct You could correlate this concept They’re saying that we will have something done on an image will shrink it and then, later on, will pass it We will again recreate and pass it to ours.