DEEP LEARNING MEME REVIEW - Episode 1

but what they haven't done names before no don't you have a show on YouTube when you review means no you had is that I think that's an entirely new concept now we're gonna steal this concept compute e five okay but first actual mean review deep learning fees welcome I'm joined by Antonia who is a bit of a meme stirrer himself and today we're just gonna look at deep learning means nice let's jump in so oh no that's a paper that is cold okay being a DL researcher is not stressed at all that is incredible how he says like that now oh I already I always knew what he worked of course yeah there is another way there was no AI winter engine is so cool yeah alright next meme nice meme I guess my brain is really big Oh what else is really big I thought you never asked every reading tough date on the edge of a really stink big gradients are always good I mean look at that why wouldn't you want to land over there that seems much more interesting than down there perfect I guess it's an O minus 7 over 4 Wow it's 10 to the that's a best small excellent the very small absolute yes hate the scientist he sees a new this is the old days yes of scikit-learn it still works be wealthy now we must use the plural for everything the next name please I don't know this template is a cool template yes and no to your researchers Bert and then exhale net what is xn that so XL net is birds just train differently okay and it costs like 10 times more straight okay and it's a bit better how much does it cost electricity y2 so people have calculated this to Train one Excel net costs about 200 DK this insane but as you work 1% better it's like that is like five PhD students that's almost the scooter language model as Excel net I know how much is better than dirt a bit Oh a bit that's all Wow state-of-the-art search our type for preprint search github for code ask random idiots on Facebook in some ways actually it is simpler to publish something on archive and not being completely like people just saying stuff like that because they're probably good unnoticed probably gets unnoticed right yeah no Facebook it doesn't get unnoticed yeah that's you have a real Q review exactly if you are in a very good mean page on Facebook about deep learning you're gonna get rekt yes exactly nice that's not gonna happen this software engineer design a chat board to chat with his girlfriend why he's busy at work however the girl eventually got suspicious over the speed she was receiving messages from her boyfriend modern problems require modern solutions they're also like pretty good chat BOTS that's just suspicious with the Piney help order for the actual content well fashion companies try to sell us what we really want fashion and mist fashion Emmys this is the new cool thing so cool does anyone use it by the way I found a huge saddle point nice mmm is that oh it is not very Nemus where is it how much accuracy do you get on fashion like MSM this is so easy like basically a doneness I don't know I'm not a fashion person so I don't know what to call this what move it no this is a pants with the boys after using drop out oh I don't know where they come from where do they come from this no it's some comic they are so beautiful are you still watching

machine learning tutorials on YouTube do you check my internet history why can't you watch porn like a normal tire and drew Angie what is this Andrew ng I must use more Kerris code yes please what is wrong with you cuz Andrew ng boy I know sir but I understand you're respected in LA he does he says it's okay if I don't understand everything yeah we're as important it's completely different not okay wait what was the plot why when your binary classifier produced with the 1% accuracy it ain't much but it's all that's what you want to get random exam change your random seed until you get 5,200 if their method works yes exactly and also like you know buddy finance but actually state-of-the-art right in work in finance reduction of the last you have a profit on things if you predict the last night next time point as the new as the last time for that's probably the best thing you can do I'm going to switch my PhD topic yeah they look like some people with their fancy methods do works because of this embed and just because it's just like you just predict whatever was very new good okay next me next - better than using double I will start with you know this math MATLAB big learning tool box yeah recently they introduced new neurons neuronal stuff like with the networks yeah and the graphs which is basically as the brain yeah and so basically you can learn stuff with MATLAB exactly Wow exactly can you learn to uninstall it I look like all you need go look like an idiot I dunno that's what we really want exactly me I sure hope my models error rate isn't super high and sorry optimization is hard yeah it's all you do is advanced methods and then there's SGD yeah it beats you every time yeah that's better me and the boys about received the Turing award so fancy yeah look at them it's probably thinking about capsules yeah oh but wasn't it like two years ago yeah what is the fate of all this is still thinking about it okay I didn't get what happened well they sold out for different oh they're different yeah okay so yeah they're not like the same ah so that means that they were kinda another way yes but only do other things sort of but can they do it on the same task ah I see no they're like trying to abstract concepts into these capsules and then the capsules can route the information to other capsules yeah does he work right kind of we kind of works yeah uh-huh who's why are people okay like you can make it do something okay capsules and that's me my desires are unconventional so r-tx 2060 2017 and 2018 so bad like use a transformer in lsdm hey I failed you know Ireland's must come back yes exact or incomplete not assistant remember this location okay I remember that what did I ask you to remember I remember what you told me this location what does this location means these are the top results machines are about to take over the world this is this intelligence yeah exactly

yeah we must be very careful also with jobs and stuff what what you finish the meat not yet there's one more so I have to preface this so basically this is a so the robots supposed to get the ball to the target and in one setting it has a reference motion of the human doing the same thing yeah so it learns to learn from that and then there for a comparison there is no reference motion and it just learns from scratch so first is with and I three times and then without with reference motion nice and now without yes we are a I do yes Donald damage is done yeah I mean I can see an army of robots they just take the bullet all right this was episode one of deep learning me review thanks so much for being here with us and have a good time