| 00:03 | So now we start recording, and we'll have to use your arm a |
|
| 00:13 | . You see my screen? Eso the lectures ever is like I |
|
| 00:19 | , or so already. Pulse um, Piazza and I give you |
|
| 00:25 | see them there. Right. So you're Steve. This okay? Someone |
|
| 00:32 | Before we actually started today stopping, , I posted this. Love |
|
| 00:39 | We'll get PSA. Uh, the a dozen in eighties that I drove |
|
| 00:43 | assigning originally I was blood is enough a short films and one final, |
|
| 00:51 | , giving that we wall Sinjar week things that weight one snow, then |
|
| 00:59 | assigned. So we're going to be if they still t short minutes Rogers |
|
| 01:05 | then one point processing. And who's your final closure you're raising on? |
|
| 01:14 | selected to go to the lesson thing the social distancing is making your |
|
| 01:20 | It's okay. Tubal. So And I don't like commission there. |
|
| 01:25 | in half often. It's just because this class it gave. The most |
|
| 01:29 | thing is is fewer your health and else itself as well. I still |
|
| 01:35 | you to feel fresh air that you to, um, need for the |
|
| 01:41 | . This is okay. Yes. . Um, are you still hear |
|
| 01:50 | okay? Yes. Then you gave Yes, and can't hear you. |
|
| 01:57 | , because I have a full screen in order to see the this |
|
| 02:03 | I thought I could still see you . Did you have a check? |
|
| 02:06 | I think I can have there should be a sound if someone pulled something |
|
| 02:10 | jumping can see it my own. guessing since clinic, you can put |
|
| 02:15 | in on your conscious. You can a new yourself and say that something |
|
| 02:20 | to you. Is that okay? So over, didn't you? The |
|
| 02:27 | in the syllables. That's anyone have about that know which is okay. |
|
| 02:50 | don't think that is No. so I think that's it. I |
|
| 02:59 | it's just the one. And I'll you what's coming in the next |
|
| 03:03 | Once we're done with this. So up before today is good. Extra |
|
| 03:10 | and beddings. So what we want do here? Iss, um, |
|
| 03:19 | review very briefly in the real English , So those are the very simple |
|
| 03:23 | where you just stood on people when in there week to do, |
|
| 03:28 | the relation of the next word was to leave. You guys remember the |
|
| 03:32 | place and family with models where we predicted by some frequency itself off over |
|
| 03:40 | few months is writing or causes. I was very the word that is |
|
| 03:45 | complex. Why, this is a idea. And how about, |
|
| 03:51 | some people have experiment? Didn't which is one of the condition wise |
|
| 03:57 | vengeance. So we worked during Normally work before, you know, |
|
| 04:04 | doesn't mean much modeling this this went a diet or you can do this |
|
| 04:09 | actually already done by kingdom. And student. And so if you just |
|
| 04:18 | divorces of people are on Enbrel, like that. You never can predict |
|
| 04:23 | the that will be a result of , right? Because the urine it |
|
| 04:29 | , say sometime, um and you also be very nicely with respect to |
|
| 04:37 | amount of turning data. And it has it is is that that you |
|
| 04:43 | only see a fix window of you know the duplicate. So perhaps |
|
| 04:47 | gets in the 1st 3 wars and predict the next one with four |
|
| 04:53 | but is doing very link. That still remains six. We bills yourself |
|
| 05:01 | . It's so the Requip. You , when there was said what introduce |
|
| 05:06 | me before in our last lecture. it was before this whole thing has |
|
| 05:15 | . And so you guys remember, , are in a is, |
|
| 05:24 | you know, he wasn't finished. Point. That is your age. |
|
| 05:28 | I from player, and then you that to predict Thistle knocks out. |
|
| 05:36 | . So the very dog layer behind probabilities did not use my four inches |
|
| 05:44 | us a little bit. Our heart aching. That point is a screen |
|
| 05:50 | of it again. Thank you. . There you go. Okay. |
|
| 06:01 | . So this is a human Um, this is the last state |
|
| 06:06 | in their network. You can use , you know, in the year |
|
| 06:09 | then, um, generates a mice over good, desirable cavitary. But |
|
| 06:16 | is going to be so. It look like this being produced in sequence |
|
| 06:22 | words, right? It's like, , we have at the bottom where |
|
| 06:26 | work and given daily or inner and upward is the cheap. That |
|
| 06:32 | Um what's so he went to Is that service on sentence and |
|
| 06:38 | For what? The Iron Municipal, Project is a work, I Then |
|
| 06:43 | you keep it in this war, spending on your to predict. So |
|
| 06:46 | so in support. Um, either when you lived in all senses in |
|
| 06:52 | flash processing, right. Um so But we can see that the |
|
| 07:01 | the loss of the the last professional on then is there some of the |
|
| 07:08 | locally Who's an inch position? So every position you already know the gold |
|
| 07:14 | by this, Like, you they didn't get the next. That's |
|
| 07:18 | you know, measures up backward. is in the network to see me |
|
| 07:25 | during pretty didn't explore. So this something you can do. Like I |
|
| 07:31 | , you can friend mint your equal into batches there is in size is |
|
| 07:39 | can't abuse thumb in a more efficient . But he says here, six |
|
| 07:44 | anything. You give it in battle write all the first words for your |
|
| 07:49 | . And then pretty soon it's one you give the second war pretty |
|
| 07:53 | is what someone is the fourth. of them in addition's off these type |
|
| 07:59 | new. Anyways, our linguist morals this, um you have a wellness |
|
| 08:06 | rare words. What? You're in husband saying words. It's gonna be |
|
| 08:10 | hard for the little tube You to great kissing was not of serving, |
|
| 08:14 | set. And so they cannot just of a degenerative position from those |
|
| 08:20 | So there is some recent war. trying to phase that for all of |
|
| 08:25 | . They do something for, the winter. Avoid your mechanism. |
|
| 08:32 | pointed mechanism will, um will uh, somewhere predicting recently of servers |
|
| 08:42 | your input. And you will also that words is where this is in |
|
| 08:47 | single. Even if you didn't see , a bring a sudden, garish |
|
| 08:50 | general, even last injections. So that's something we can consider. But |
|
| 08:57 | is not the best models that they some of these buildings since irrationally Not |
|
| 09:02 | was thinking too. So it It'll be off. Respective war beddings |
|
| 09:07 | contextualized. That is So how are zipper? From what we have this |
|
| 09:14 | pages, you have already talked about one in Venice resemble work today Plastics |
|
| 09:23 | little. And the girl with this models is you guys remember they did |
|
| 09:30 | know that was to sell. The it is insinuation problem because they learn |
|
| 09:34 | six, uh, warmer presentation for voice, um, in them use |
|
| 09:41 | , you know, he gets one representation. Even if you drink a |
|
| 09:46 | of those a day, you should investing problem that this sense of the |
|
| 09:54 | was to the Oakland. But until single better and like condition before the |
|
| 09:58 | frequent sounds is going to going, the British. So then is the |
|
| 10:05 | of Elmo actually decided to fix his , and the medication is that you |
|
| 10:12 | be able to use a contest. so that's a water presentation. Gina |
|
| 10:19 | in the pen in from the older that racism so and tries to do |
|
| 10:26 | . So you see this diagram here it's trying to have. It's a |
|
| 10:31 | that when trying to predict play, this conflict a high percentage in for |
|
| 10:37 | war play even this country, you to a degree recitation than our visit |
|
| 10:44 | for the same way. But I'm these different contexts wherever the second contests |
|
| 10:50 | May with the new medicine so cedar like probably show and the prison reverse |
|
| 10:58 | actually again doesn't still want to Please let me want to do. |
|
| 11:06 | were doing language. Little's use didn't expeditious, but they wanted it in |
|
| 11:14 | day. Supplies. Let's see how do. Um, key ideas. |
|
| 11:24 | want to use it on the wish . The context of award. |
|
| 11:29 | So this is in a little gum of anguish models. And it was |
|
| 11:35 | beautiful world and our noticeably so before A context before you. Perfect golden |
|
| 11:43 | . But her What they say here what if he has your diet |
|
| 11:47 | Nothing prevents us a trying to do bidirectional in with a little and actually |
|
| 11:52 | from the very end all the centers believe the biggest work. It's local |
|
| 11:58 | both of her right. But they do it. Remember within see how |
|
| 12:02 | would work. So the architect er allows them a little too close to |
|
| 12:06 | Z and betting by stones in so words. So here's what we |
|
| 12:13 | um I I important wishes, So moistened deep architecture. Guess he has |
|
| 12:21 | lot of years. So let's the of this in time. But what |
|
| 12:24 | have here. So the orange the that is in first layer? |
|
| 12:30 | what was the the wisest? That's signature, for this isn't the most |
|
| 12:33 | the morning, so the words air into combinations of Carter's. So the |
|
| 12:40 | layer it is that a conclusion under the airway said he's taking guarded Raise |
|
| 12:45 | presentations for the port So the word straight into a combination Carter's. And |
|
| 12:51 | he have as a voice in his is E. The things that used |
|
| 12:58 | drink memory representations needs the war that have a joint set we consider go |
|
| 13:04 | some. No relation is Richard because we came back up to the |
|
| 13:08 | representations. And in addition to that , Luna character of representation of supposed |
|
| 13:16 | wear board to back passes, fast that they just take the single cycles |
|
| 13:20 | crackers as a whole means that, , you can liver's, um, |
|
| 13:27 | information friendships and was learning about the . So the operative, the CNN |
|
| 13:36 | going to be bad into a bidirectional . Um, so the 1st 1 |
|
| 13:42 | grisly that was, um, the the same. Here is the |
|
| 13:46 | um, LSC a model, then next sequence in the honesty, |
|
| 13:50 | hum, it's the I know this just what was that for? But |
|
| 13:59 | he has a layers Elysium here, , and then that will generate at |
|
| 14:05 | representation for it's for. So we go the worst all the way |
|
| 14:09 | right? And then we can use representation can predict the next word so |
|
| 14:14 | see play. We can use your presentation to protect the morning |
|
| 14:20 | then the Lakers and the intuition is we'll have to We need ourselves to |
|
| 14:27 | the forward and see um, we go by direction and in a similar |
|
| 14:32 | we give us, it's the From the very end. You see |
|
| 14:35 | same representations we have exactly the same staying in later. And then your |
|
| 14:41 | and years for resigning. The A little hearing. This is a |
|
| 14:46 | playing with the dates. Okay, the same process is a little is |
|
| 14:55 | same architect er for these Gray was model. And for this one thing's |
|
| 15:03 | visit, number of milieu is the number. And, uh, so |
|
| 15:09 | exactly the same. Okay, um, now we have we have |
|
| 15:16 | information right for the work when I work. But if you avoid the |
|
| 15:21 | things, we can use the last of the model to pretty swollen warm |
|
| 15:27 | aggregate the Brazilian work, right? , very mild. He's actually |
|
| 15:33 | He said the learning objectives. He the same before the other buddies were |
|
| 15:40 | his moral architectures to look us that to predict that you used the bears |
|
| 15:47 | the next one. Whenever he uses way. Well, because isn't that |
|
| 15:51 | right in the middle? Just use little presentations for that. But so |
|
| 15:55 | we have to We have many choices do because n has actually many the |
|
| 16:02 | . It can reviews a movie you representation for the actual war. So |
|
| 16:07 | don't care about the prediction we only as a learning objectives. The |
|
| 16:12 | But Emerson's the model day. But don't care about the steps along with |
|
| 16:18 | about how good are these representations? need your presentation that we have |
|
| 16:22 | right? And so then we have choices on how to do this. |
|
| 16:28 | we get big the last year we're to Disney, you're here, or |
|
| 16:33 | can say the C in earlier we take them. All right, which |
|
| 16:38 | what they decide to do in the picture is presenting tell. So in |
|
| 16:44 | end, they do. Is they being the only the native presentations the |
|
| 16:51 | of the last year, the major of the Great either. And then |
|
| 17:00 | is a patient comes in a car then the you visitation for everyone. |
|
| 17:08 | because when you're generated the gym person very possible alone, all of the |
|
| 17:13 | in the sentence And the base, course, expected that these years are |
|
| 17:16 | to be different. This is not . This is not gonna be different |
|
| 17:19 | . It is always the same war the only the bends of the doctors |
|
| 17:24 | , xyz, the same work. then we'll see there afterwards is once |
|
| 17:29 | beautiful day in the conference. Let's come Texas, and that's that's |
|
| 17:34 | Actually do something more sophisticated. which is that they actually introduce some |
|
| 17:43 | up in ah, ***. He's weight toe age of these representations. |
|
| 17:49 | basically, what is doing is that each for presentation? You know, |
|
| 17:53 | , I don't know, weekend to was how much and mornings 30 suspicious |
|
| 18:03 | maybe for a basket is much more addition to the C in and out |
|
| 18:09 | it in a year in the last . And what they recommend is that |
|
| 18:14 | not coming up with, um, a little tickle. My basic actually |
|
| 18:18 | to be fine, too, for is. So, for example, |
|
| 18:23 | we want to use an Emery presentations you run in the recognition? Then |
|
| 18:28 | take it between Elmo on the Don't talk. You're a days a |
|
| 18:36 | to make predictions. Or Syria, one works better for for any are |
|
| 18:43 | when you dio is that, um you're bringing your CRF people forget the |
|
| 18:49 | back. And when you went in sound uprightness going to allow the model |
|
| 18:55 | year and how to combine these you . So he's been there. A |
|
| 19:02 | of your down downstream. How is ? Is everyone still there? |
|
| 19:16 | it's here. So let me ask a question about this check. So |
|
| 19:30 | you explain again about for why, could we have one layer of CNN |
|
| 19:36 | layer off the end of last Yeah, so that's the definition. |
|
| 19:42 | actually, so anything Exactly how Um, ministry. We know exactly |
|
| 19:48 | many they use. We have some of Tennessee and can have a fusion |
|
| 19:51 | use inside of so limited in So the infusion it is orange. |
|
| 20:03 | , representation is very stolen. ulnar networks. And so they |
|
| 20:10 | Last warning or average pulling over um bygones tried and forget. So |
|
| 20:17 | think holder presentations actually go toa very end, naturally lens of the |
|
| 20:25 | But it's something that gait or something that that is much fired. And |
|
| 20:30 | we do traditional legal conference and the is the the generates representations from a |
|
| 20:38 | and then for all your body goes the world. There were already tried |
|
| 20:43 | in there for all the organs. then in the end, they do |
|
| 20:48 | really good mass pulling on all of . Those presentations to end up with |
|
| 20:53 | single character, raise your presentation for world. And again, the idea |
|
| 21:00 | for very intuition and say they want allow the model to her stomach. |
|
| 21:04 | logical information, A lot of words , this is this is very more |
|
| 21:11 | , and it will be a face for the words. But then that |
|
| 21:15 | going to be starting to be a cm. And so the bidirectional model |
|
| 21:21 | the calcium. So the actual Ellen , um, will see the sequence |
|
| 21:25 | words and movie able to generate a for the words that is taking contests |
|
| 21:32 | out. But it's a very that model some way. Um, |
|
| 21:54 | No. I just have the dimensionality that the mission and it is so |
|
| 21:58 | presentation of the car or whistles. they're the ones that are they have |
|
| 22:03 | Imation's, each of them, so highly the additional for the importance of |
|
| 22:09 | sack. Size goes, I need . And so, um, this |
|
| 22:24 | a big architect. Er off the . I don't think it's well explained |
|
| 22:30 | the paper s So this is just diet that says you have a non |
|
| 22:35 | , um, here that the same sophistication. You haven't heard from him |
|
| 22:40 | a year. Knowledge while it ends a lake. You know, this |
|
| 23:17 | going to help him. We'll Yeah. So I didn't any other |
|
| 23:38 | . Oh, fine. Animal use , Ellie was do vial is, |
|
| 23:44 | , layers. Yet with 4000 and students and 512 and mentioned gradations. |
|
| 23:58 | a very, very low. Anyone else has other questions. |
|
| 24:16 | Yeah, it's just a very Right. Um, they do a |
|
| 24:22 | thing completely before. For example, you're trying to do, uh, |
|
| 24:28 | and Syracuse patient so you do the in completely your training, and then |
|
| 24:36 | do the CFR, for example, ? Her. Okay. I just |
|
| 24:41 | to verify, like you're not saying it fails to be doing the CF |
|
| 24:45 | and also turning the Elmo, No. Okay. Yeah. |
|
| 24:53 | So the the 1st 33 animal. ? And remember, that is being |
|
| 25:01 | . Support training. Amble. You need this final presentation race long into |
|
| 25:12 | . This is an alarming to, know, do that. I want |
|
| 25:15 | use, you know. So, , you don't need this objective. |
|
| 25:22 | playing your Elmo model is your and my model is straining. Predicting |
|
| 25:28 | left or right the learning or getting this model isn't to protect. The |
|
| 25:38 | were in the next one, Every lady just need up workers. |
|
| 25:44 | harsh shortness. Um, And with very large, Corp, it's you're |
|
| 25:52 | to train all this grammars Write the for your CNN, the problem through |
|
| 25:59 | elecciones and and what you generate does from eating representations. Then once you |
|
| 26:05 | that, you're going to us. gums don't drink us. So the |
|
| 26:11 | write stories emerging out of here um, layer, Um, a |
|
| 26:18 | later, before work during work or , It's your back. And and |
|
| 26:24 | this here is it strained to do are so basal insulin. They're in |
|
| 26:28 | of the and you are you Propagate here for the promise of the |
|
| 26:32 | and that we're forgetting here to the on how to combine this facing kidney |
|
| 26:38 | presentations. So when you say why you need can't use the same |
|
| 26:46 | document you're trying to do. Um name and signature in on Can't use |
|
| 26:52 | same document to train to stop the . You know, uh, to |
|
| 26:57 | the l more like, Why do need a another cop was bank to |
|
| 27:04 | Elmo. Okay, so you need not that you need another one, |
|
| 27:09 | prevent you from using the same corpus same train set. The problem is |
|
| 27:17 | , um, this is the very model. It had billions of |
|
| 27:23 | So you're going to need a very corpus. This is per vice. |
|
| 27:29 | know, you don't need labels for , that you need a very harsh |
|
| 27:33 | . It's a in contrast for the that we care about our our |
|
| 27:39 | They actually pretty small. Is the early enough to where he didn't go |
|
| 27:44 | any? Okay, that's why traditional always do that. You take a |
|
| 27:51 | dream all whether it is or birds , um, the GPT or, |
|
| 28:00 | , your know or plastics, They train because they haven't drink a |
|
| 28:08 | chewable is a beta. Then you use it when you're down to found |
|
| 28:12 | cats. And you can buy a to go to, like, |
|
| 28:19 | So you can also add the documents trying to do like any are |
|
| 28:23 | Right? In addition, Is that ? Yes. So our study and |
|
| 28:31 | is what we call it's fine. , take a deep brain will and |
|
| 28:36 | find tune into your past. Meaning Corp. It's even if it's not |
|
| 28:40 | in here. But you find any your data sets and then your brain |
|
| 28:44 | the and desk as well. But can do a single dynasty, a |
|
| 28:49 | . So when you do is, it will talk about it. You |
|
| 28:53 | . You're not doing fine. I , and you're just using preaching |
|
| 28:57 | and you're gonna freeze the promise for . And you only tried Provigil related |
|
| 29:01 | your down strength us. But if doing fine to me, then your |
|
| 29:06 | race Elmo use you adjusted the numbers your data. Okay, Sounds |
|
| 29:15 | Thank you. I do notice that is working. Did someone else has |
|
| 29:21 | question? No. Think it Okay, so, um, |
|
| 29:37 | What are the advantages that we have ? And here's your looking at Ms |
|
| 29:41 | , right? Yes. Yes, can see your screen. Okay, |
|
| 29:53 | you see the little broker piece slide ? This so carpet isn't that it's |
|
| 30:02 | supervise, right? It's training on muscle data, but it only needs |
|
| 30:08 | sentences because everything you need to train presentations is in there. It's just |
|
| 30:14 | secret supports, um, is interest sexual presentations, which is the meaning |
|
| 30:20 | that ditch in where we were trying so go back through the biggest |
|
| 30:24 | They were fixed, the Gardasil, comforts of the words. And we'll |
|
| 30:29 | you a, um, a different and ready for, for your words |
|
| 30:36 | . Or West was surrounding that particular . And she's much more, |
|
| 30:41 | convenient. I guess it's ever find presentation for your ward, right? |
|
| 30:46 | a deep model. It has a of layers, system me a lot |
|
| 30:50 | parameters in his character base in like that has the advantage that he announces |
|
| 30:58 | do much more things with with this the other models. In between birth |
|
| 31:04 | , when we have used for, , next language data, where the |
|
| 31:12 | themselves are we seeing more feeds, so when you have words from differently |
|
| 31:18 | coming together in a single ward, have better models and your all the |
|
| 31:22 | , and I'm gonna work because they work at the more level, and |
|
| 31:26 | we're gonna have your presentations. Where wars. But when? L know |
|
| 31:30 | can actually use the press addition something the Angus. Then make up the |
|
| 31:37 | . And in versus because you can plug it into any of your done |
|
| 31:43 | tests. And that's what people have , actually. So when this came |
|
| 31:47 | on, our people started plugging that . But presentations and they have performed |
|
| 31:52 | we have seen, um, before sitting here It is, um, |
|
| 32:00 | . And why weights? Um and is your some experiments. It |
|
| 32:04 | He offers, Have that. And there's some evidence that this is, |
|
| 32:08 | but this is right, but it's cracked and does 100% whether that is |
|
| 32:15 | has happened in in the model or . So the linguist models a lot |
|
| 32:20 | nursing techs enzymatic knowledge. We know because you have seen that in any |
|
| 32:26 | watch model, right count base, on English models also learned, see |
|
| 32:30 | not acknowledge. And so it is reasonable assumption to say that, |
|
| 32:38 | they will do is that my own can. All right, so |
|
| 32:52 | the dealing with lousy and to him to this. I was representing and |
|
| 32:56 | to this. I wish your presentations use. That's also very nice. |
|
| 33:00 | remember how we have the lamb that in the land of parameters work the |
|
| 33:05 | . How much relevance we're going to through the character representation you can meet |
|
| 33:10 | presentation or the last layer is that is very nice. Basically, you |
|
| 33:16 | your end's model Kansas. I can shoes, right? What is more |
|
| 33:21 | for the dust? And that's what missile show you here is. Here |
|
| 33:25 | an example. We're trying to keep apart. When Elmo is it's I |
|
| 33:30 | . So they will find on the has, um worse isn't situation |
|
| 33:37 | They will. Six on the right the bar speech died in past. |
|
| 33:42 | so the hers three. Our state the art the first uses and represent |
|
| 33:50 | yet measure of the state of New . Then they're comparing against collector's on |
|
| 33:56 | , and they're still model the V list. Um, the first layer |
|
| 34:03 | the second the your elbow. And so what they're trying to do |
|
| 34:08 | , so I do these apart? kind of knowledge is this? Are |
|
| 34:14 | still years representing? Is it most tactically on the Inderal presentation or is |
|
| 34:20 | a most romantic Leonean other presentation? what they do is that just looking |
|
| 34:26 | there the repetition of the first And so that's what we see |
|
| 34:31 | Violent person here and violent person. here, Um, and they give |
|
| 34:36 | to, uh, linear. people never version production system on and |
|
| 34:42 | they think the second layer and irregularly ablation system on day productions so we |
|
| 34:49 | see here that reverses its ambulation to layer for em. Kids better |
|
| 34:56 | And it is very competitive results compare state of the art. But that's |
|
| 35:01 | you do it and even our you know, progression system, a |
|
| 35:05 | system and you can get very close wasn't this initiation, but it will |
|
| 35:14 | at our spirits dining table sets. first restarting the first layer is the |
|
| 35:19 | that a chief seven was so through sick earlier. All of here, |
|
| 35:23 | know, the results are closer. there's it's a little little harder, |
|
| 35:28 | . Son was in its in its configuration. It's clear that, |
|
| 35:34 | other learning different thinks you and that have different bristles, right? It's |
|
| 35:39 | we can clean it with us impacting leisure. Maybe that's still much, |
|
| 35:43 | they're pretty kind of specialized, you , different that religion. That's what |
|
| 35:49 | advantage of having three presentations intto wonder you can never say years. |
|
| 35:58 | another analysis them by the office of animal paper on they. They're claiming |
|
| 36:04 | the model is very efficient in the . The brain is that this is |
|
| 36:10 | center of the and task, of , that things that you're you're pretty |
|
| 36:15 | that needs that you have to. puffs took hers in each foot. |
|
| 36:22 | is the end all in since the know and I don't want. The |
|
| 36:26 | nine is a basement system, and is present the train set getting, |
|
| 36:32 | know, skip and so you can how, with less a month of |
|
| 36:38 | , the animal model is actually able learn faster. So it's getting better |
|
| 36:44 | with less amount of the drug, then here in, you know, |
|
| 36:49 | . What's that? And it started any. See friends a look. |
|
| 36:54 | in reality, animal has already been a lot more days. Right now |
|
| 37:00 | , um use. Um, I said that it's nine and |
|
| 37:04 | so I kind of feel like these boys in the cell. What? |
|
| 37:06 | be okay, Okay, because of between isn't super by, so it |
|
| 37:13 | cause that much. I said it's your price in anyhow, So because |
|
| 37:22 | before, I'm not gonna be integrated almost any little and I'll be task |
|
| 37:27 | can choose a stable look at the , right will be in bed. |
|
| 37:31 | then you're into your presentation for the when you can just don't use, |
|
| 37:35 | just as they are. No, don't getting it and just add a |
|
| 37:40 | toe to do your last election. then you do the one of razor |
|
| 37:46 | ask. Now there's There's an interesting that explore want to do with him |
|
| 37:55 | these older models. Respect to find need or not. Fine to |
|
| 38:08 | Bye, Alan. A. I that baby off people in the condition |
|
| 38:18 | just so they talk about proceeding daddy's protein and then is your way. |
|
| 38:25 | you remember the waste of the task predicting what you want to do. |
|
| 38:32 | you keep all the other parameters right? So you don't want if |
|
| 38:36 | get into immediately your solution and so fixed. The only thing the model |
|
| 38:42 | use animal for urine test, that's they call a pro. Seven |
|
| 38:47 | When would you know my doing? are back with about in here all |
|
| 38:51 | way to deal more parameters. We're for me and task. That means |
|
| 38:56 | in back changing it. But I'm animal model, and they compare different |
|
| 39:01 | . Um, here's and missiles results using frozen animal. This they're different |
|
| 39:09 | that people have used to compare, , deeper systems. Some of them |
|
| 39:15 | semantical labor being Reppert resolution and density . Um, I think he's endowment |
|
| 39:26 | my doing, which think so. and I think this one is He's |
|
| 39:34 | question answering task because this is a state of the art for, |
|
| 39:40 | for that task. So the were that on the paper was foolish. |
|
| 39:45 | is the venous system. This is animal blast baseline. And so this |
|
| 39:50 | , you know, the little system you are element when you get better |
|
| 39:54 | souls. You already improve over state the art in all cases. |
|
| 40:00 | But you're not making any medications to was just done in animal to your |
|
| 40:05 | system again. Um, who is ? Is also using the's fine. |
|
| 40:15 | . Name is you haven't Here. Okay, Um uh, one of |
|
| 40:28 | my slice when it would need to , well, the biggest size. |
|
| 40:38 | this is what they find in the , Um, after many experimental |
|
| 40:45 | and I guess it would be consistent one birth. Birches the just for |
|
| 40:50 | amazed and many that we're gonna Um, do is for now, |
|
| 40:56 | think that's the joke is coming after one. A. So the eyes |
|
| 41:02 | can mean that is frozen in the need is fine. Tune is the |
|
| 41:07 | says here in conditions is between any , there's no it's frozen in for |
|
| 41:14 | drastic islands. Are many, that's the numbers. A needle test |
|
| 41:20 | having dinners, and, um, is this year? It is a |
|
| 41:26 | specifications best. Whether it's any of trippy drain, it says you're visiting |
|
| 41:33 | the president process But I understand by many things have similar performance. He |
|
| 41:38 | a sentence there. I, for , would be in jail meant |
|
| 41:43 | It something like deciding. Tell then for any other ligamental do |
|
| 41:48 | Um, but I'm Marisa for bed birth. Their community views planting because |
|
| 41:54 | sugar results in this regiment, I still, you were deciding to use |
|
| 41:59 | is that it's a good for To keep an eye on this paper |
|
| 42:07 | this is in case you need to that. Here's a simple, |
|
| 42:16 | even paper. And I started doing nice Why this idea of crickets in |
|
| 42:21 | night? It's more the biggest work it? It's good. It's actually |
|
| 42:26 | necessarily answering the why. This is good objective. It's just showing that |
|
| 42:31 | is reasonable. Um, it gives bed wrestles. I think I want |
|
| 42:36 | look at the reds. Occur. recorder is an English model, the |
|
| 42:42 | English model. Um, this uh, both starting accuracy. More |
|
| 42:51 | pass under meaning initialization. We have . Um, I mean, this |
|
| 42:57 | good. Really simple, but a blanking on what is the empty in |
|
| 43:04 | tea? Me that's in should take listen. But the important message is |
|
| 43:12 | the girl is, you know, learning. Objecting over the language model |
|
| 43:15 | the best results this attention mechanisms. actually going to learn them on |
|
| 43:21 | Where did she needs and have to a pension? Corrigan. It's |
|
| 43:26 | We don't call that. This has a covert. It's next. So |
|
| 43:32 | summary for today, um, Fessler are suboptimal because is the face with |
|
| 43:40 | country before right? They cannot come a maturity. Um, so is |
|
| 43:45 | is not ideal extra large silvio. hands allow us whenever's a sequence. |
|
| 43:51 | so it's, um, less amount the organs of the presentation of It's |
|
| 43:54 | Doctor. Kruger Specific Corp is the you're getting much models. Reason that |
|
| 44:01 | working in the context of those, , also nerve burning diesel backs of |
|
| 44:04 | ricin can dribble was on knowledge. was the very first of all of |
|
| 44:09 | will normalize your games and just and general, the condition allies representations are |
|
| 44:17 | weaking museum to improve the cells in tests or whatever. Really, doing |
|
| 44:22 | that's, um it's a good idea throw in some some of this |
|
| 44:27 | Nice representations, Really big abscesses. that's it. Um, I didn't |
|
| 44:35 | any more questions. And you guys questions. Are you still there? |
|
| 44:52 | . Yes, yes. Any Yes. Okay. It looks like |
|
| 45:17 | your memory's maybe well, some I , but you and has questions. |
|
| 45:36 | you got a posted? The recorded ? Yes. I'm gonna pose it |
|
| 45:42 | Medio in the portal. And any questions? Yeah. Are you guys |
|
| 46:10 | with, um, doing sleep? off your face. Shooting the a |
|
| 46:21 | of this, right? Yes. why I'm asking its way. Comfortable |
|
| 46:35 | that. Okay, so thank guys for very With me here. |
|
| 46:45 | a little bit. Um I'm a bit stressed doing this because I'm |
|
| 46:48 | This is not a traditional setting. just the messy in your bases and |
|
| 46:54 | reactions, cause I can add seen lice in your face on the same |
|
| 46:59 | . Um, I welcome any comments be five. You want it to |
|
| 47:05 | , Uh, be a pizza. can check it. A committing that |
|
| 47:08 | also. Too bad a comic Make sure you just get that because |
|
| 47:12 | easier for me. Like I I have a special K circle that |
|
| 47:16 | . Besides best like class and the to anybody can come around. |
|
| 47:22 | do you think of mentioning that easily permitted? Got all these hours like |
|
| 47:27 | used to. I'm going to still to be really in our initial |
|
| 47:32 | but I think it invested. We a poster just being on the |
|
| 47:36 | looking for a job or lining in physical things right. I think I |
|
| 47:42 | be the best way to go to with Serious that you may have. |
|
| 47:46 | mean, he's very full life So just do that Baby said, |
|
| 47:54 | , I have a question. What in or meeting you get? It |
|
| 48:06 | This is really the most side, starting this. Also, I have |
|
| 48:13 | case at home and with my with work and it's in. It's pretty |
|
| 48:20 | you guys also cases we've been So so excuse me for weight loss |
|
| 48:36 | it could make the best out of . But I still want you to |
|
| 48:40 | out of the facet. Some useful visual movements don't bigness that you can |
|
| 48:45 | you 30 years. So state social, decent thing is resulting to |
|
| 48:59 | that this time. Okay? And no questions, So just wondering if |
|
| 49:10 | would extend this semester. Yeah, think I think you would be hard |
|
| 49:23 | this, Lee, but I don't it's a reversible thing to try to |
|
| 49:28 | like students. You guys getting, know, talk to the care. |
|
| 49:33 | ? So this is something that I not decide, but you still feel |
|
| 49:39 | , you know, having a discussion the chair. You been serious about |
|
| 49:45 | these will be back to yourself? think we can salvage it. And |
|
| 49:52 | has so much to work in the of Jonas? Assume you have the |
|
| 49:59 | to I think that that's a discussion , um national wife by skin the |
|
| 50:17 | listening. But But I know that is a very unusual situation. |
|
| 50:23 | you know, you know, was the normal here? Anything else? |
|
| 50:42 | to exercise? You know, I if it's indoors, I mean, |
|
| 50:48 | a noise. I think that would this little bus. The studio plus |
|
| 51:04 | in there. Everyone needs guys from . Is they have any questions |
|
| 51:13 | Stay healthy, are you? Okay. We were We went with |
|
| 51:17 | decision. Addition. Mechanisms way. invited Lecter you Some trouble coming. |
|
| 51:26 | weigh. Have transformers. Roberts Very good. It's a good |
|
| 51:39 | So I'll see you |
|
