video_id
stringlengths 11
11
| text
stringlengths 361
490
| start_second
int64 0
11.3k
| end_second
int64 18
11.3k
| url
stringlengths 48
52
| title
stringlengths 0
100
| thumbnail
stringlengths 0
52
|
---|---|---|---|---|---|---|
qoGBO3q5ikI | up into the clot in the brain but typically it's through the vein question in the back there yes your head the question is for the statin agents does it matter if it's named brand or generic there doesn't seem to be a key difference there so I think the generic is fine far in the back yes it's inserted in the question is where as the catheter inserted its inserted in the | 3,611 | 3,642 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3611s | Stroke: The Basics | |
qoGBO3q5ikI | femoral artery and then up into the brain yes so the femoral artery is it's right next to the groin area so it's the top of the leg where you you get that blood vessel yeah so question can you go elsewhere it doesn't really take any time to push it from the leg up to the brain versus the arm and it's easier it's a bigger target it's a little safer to go into the leg than the arm but if | 3,642 | 3,669 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3642s | Stroke: The Basics | |
qoGBO3q5ikI | the leg is all blocked up you can go into the arm years ago when they first started to do this they went through the neck that didn't go so well up right I'm interested in you mentioned there are those discrete situations that they would be used in or shouldn't many coumadin people be thinking about going on to those and secondly do the hospitals in our area generally yeah so | 3,669 | 3,696 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3669s | Stroke: The Basics | |
qoGBO3q5ikI | the question about these new anticoagulants and they're there for so some insurance was call cover one versus the other the first one on the market was called Pradaxa you've probably heard about that one most hospitals would you're gonna have access to that and many will have the other ones that have come out more recently so these are typically for patients with atrial | 3,696 | 3,719 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3696s | Stroke: The Basics | |
qoGBO3q5ikI | fibrillation they have not been shown to be effective for people who have heart valve problems like a mechanical heart valve so it kind of depends on the reason that somebody's on coumadin if they're on for atrial fibrillation and they're having some trouble with the coumadin there's a you know these offers some advantages for somebody who has a stroke from atrial fibrillation most | 3,719 | 3,740 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3719s | Stroke: The Basics | |
qoGBO3q5ikI | doctors now would be talking about using one of these newer agents rather than going with coumadin because you don't have to do the frequent blood tests there's not so much in the way of food and drug interactions so they're much more convenient for the patient back there yeah you sorry I don't know anybody's name yeah so the question is can mini-strokes lead to dementia so a TI a doesn't lead | 3,740 | 3,769 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3740s | Stroke: The Basics | |
qoGBO3q5ikI | to dementia because as we said by definition a TI a doesn't damage the brain but little tiny strokes they can lead to dementia so dementia being a lack of cognitive ability so you can imagine if you're starting to knock off lots of little spots the connections between different areas of the brain are going to be affected those typically occur again in people with high blood | 3,769 | 3,789 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3769s | Stroke: The Basics | |
qoGBO3q5ikI | pressure that the small little blood vessels deep in the brain get narrowed and you start to get these little tiny strokes that people often call lacunar strokes and they can cause dementia so we don't want those to build up if there's if we can prevent a second row there go ahead yeah so question is about omega-3 and it's been a controversial area we find that the the statins have | 3,789 | 3,823 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3789s | Stroke: The Basics | |
qoGBO3q5ikI | much more evidence to support them so we would recommend the statin agents the guidelines recommend the statins over the omega-3 question in the question over there have we expired our time I think one more question okay our curricular I have friends who yeah so the final question is about these new anticoagulants and are they easier to regulate in the coumadin and | 3,823 | 3,860 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3823s | Stroke: The Basics | |
qoGBO3q5ikI | they are definitely easier to regulate they were all tested all four of them in huge huge trials where one half took coumadin and had all the regulation issues and the other half took these new medicines with no regulation okay so you weren't upping and adjusting the dose like you have to do with coumadin and all four of these you know were either as good or better than coumadin in these | 3,860 | 3,886 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3860s | Stroke: The Basics | |
qoGBO3q5ikI | big trials so they were as good or better without all that regulation so somebody who's blood thinning is jumping all over the place on coumadin these are much smoother now the disadvantage is that we don't have great tests to monitor how much is in the blood with coumadin we can tell exactly how much is in there with it with the blood tests the newer agents that's a little bit | 3,886 | 3,907 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3886s | Stroke: The Basics | |
qoGBO3q5ikI | trickier and then there's a little bit of an issue with reversal of these newer agents that coumadin is hard to reverse also but we have medicines with a lot more we have a lot more experience reversing cumin if somebody comes in bleeding but overall patients typically did better with these newer agents so I think we have reached the end thank you very much for coming I appreciate all | 3,907 | 3,931 | https://www.youtube.com/watch?v=qoGBO3q5ikI&t=3907s | Stroke: The Basics | |
2Z1B0xESzMw | hello everyone and welcome back to my channel today we're trying something a bit different which is I'm going to be talking about my research area today which is machine learning but I really hope that any of you guys that are coming from a non mathematical or computer background if you are watching along that you can understand this because I tried to make it | 0 | 19 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=0s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | understandable to a general audience and if you are watching anyways let me know in the comments below if you are able to follow along with this and if it made sense to you guys because I'm always trying to make sure that research areas can be accessible to everyone even if it's not your research area so I'm doing a PhD in computer science but my background is in maths and statistics | 19 | 39 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=19s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | and that's how I came across machine learning but some of you guys have been asking for me to do more machine learning on computer science videos and that's not something I really saw about doing on this channel but I did one to have it be more general PhD stuff but to kind of try this out for this month I want to do one sort of day a week on machine learning or computer science | 39 | 63 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=39s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | programming that kind of thing one day a week on more general PhD stuff so that might be writing or skill so I've talked about transferable skills before so it could be something like that and then one day week that's more on the personal side of doing a PhD so things like routines weekly or daily vlogs which people have seem to respond really well to and things like you know financial | 63 | 88 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=63s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | aspects of being a PhD student side hustles for students all of that kind of thing and then one day a week that's all the planning productivity stuff that tends to be what most people are interested in so that's kind of this the general plan I'm gonna have one of those each day and probably have one day of these study with me kind of videos as well but it is a lot so I'm gonna be | 88 | 112 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=88s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | trying that at this month seeing if it's something I can keep up with but what will be really important to me is understanding what is working for you guys so if there's a video style you like be sure you are liking be sure you are commenting below and be sure you are subscribing because that's what helps me understand which videos are working and which ones I should continue doing | 112 | 131 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=112s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | so if you want me to continue doing one type of video be sure you're engaging with that video type or multiple video types if you love all the videos you want to see more videos about machine learning and my kind of research or PhD stuff in general then be sure you do subscribe and that you hit the notification valve so that you know when new videos are out so this first video is going to be an | 131 | 153 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=131s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | introduction to machine learning in general just so you guys have a general understanding if anyone is like learning about machine learning for the first time here that you guys will have an understanding and then in future weeks we can get into some of the general algorithms and then as well I can show you the programming that goes alongside those things but let me know if you'd | 153 | 171 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=153s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | rather see more general stuff or you'd rather see things that I specifically used in my research so what is machine learning so machine learning is a branch of artificial intelligence which is a technique that enables machines to act similarly to humans so trying to learn how to do things the same way that people do and then machine learning is a branch of that which uses statistical | 171 | 197 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=171s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | methods to enable machines to improve with experience so they learn based on their own experience and or from the experience that's in the data obviously so machine learning is essentially applied statistics so my background is in mathematics and statistics and that's how I first learned about machine learning was in my degree in statistics and in my master's in data analytics so | 197 | 222 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=197s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | it's not 100% computer science it is more statistics you need kind of the mixture of statistics and computer science skills to be able to use machine learning and then deep learning I'm just going to mention briefly as well is another subcategory of machine learning and it's basically it has the ability to work with a multi-layer network of algorithms essentially of different | 222 | 248 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=222s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | layers of things going on and that's how it makes decisions but before I get started with talking about data I just want to mention briefly as well I'm gonna be talking about features to do with data everyone in school probably would have done some form stats where you have different variables that are the input and then you have one that's the output so the features are the input variables | 248 | 270 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=248s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | so one example I'm going to talk about later is housing prices so the features might be the rooms the number of rooms and highest the number of bathrooms the square footage all of that stuff and then the output variable will be the housing price that's associated so just to mention that briefly as well so I'm going to do some examples and I thought like how could I make these | 270 | 293 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=270s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | examples a bit different if anybody is familiar with machine learning so I decided to do things that are all based on YouTube so here we see one of my videos I'm actually wearing the same cost are kind of not ideal anyways so one of the things that YouTube offers the employees with speech recognition because you've got these captions that are auto-generated it says that there on | 293 | 315 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=293s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | the screen that they are auto-generated meaning that I didn't provide the captions for this video instead they use a speech recognition software that turns my speech into text and that's automatically done and it's not always a hundred percent obviously especially when you have things like accents like I do say that I recorded in English from Ireland and then they obviously use some | 315 | 339 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=315s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | things so I mean that part already looks but it like looks like it's correct so that's good but that's one thing that they do so that's an application of machine learning and sometimes some of these examples are also going to be using deep learning so it's hard to send a separate the two because you never know a lots of things that can be done using pure machine learning algorithms | 339 | 359 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=339s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | can also be done or are done with deep learning algorithms so the next thing that they do is spam classification so you see on my comments this is an YouTube studio but I have likely spam so they have they know when a couple of comments are going to be spammy so these are people who comment down below subversive and things like that so they get put into the likely spam or they had | 359 | 385 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=359s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | helped to review and then I can go back and approve these comments time's comments that seem completely normal are come up with spam for some reason so it's not really 100% but you know like you know when Gmail obviously you've got your spam and that's pretty accurate like I've rarely ever had anybody anything in spam that's not spam so yeah so that's one thing that they do | 385 | 408 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=385s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | that's done with machine learning as well so they learn the difference between spam and not spam by using they have examples of ones that have been labeled by people who said this is a spam email or this is an OnStar me mail and they learn how to categorize those as spam and not spam another thing that they do is text generation so you can see I have this comment here from eco wander shout-out | 408 | 432 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=408s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | to you for commenting on my video thank you so much you contribute to my growth here on youtube so thank you and and so you can see that YouTube has started doing this in YouTube studio where you have like generated comments that you can just reply now I don't use those because obviously it's better to have longer comments if you want to get the engagement up so it looks better in the | 432 | 456 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=432s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | algorithm and it looks better for you guys if I'm doing longer comments and I'm engaging with you more when I can and so these are things that they do now though you can do automatically generated and that will be they'll read the machine learns how to understand the conflict the content of that comment and what would be an appropriate response based on previous interactions they've | 456 | 478 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=456s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | seen on YouTube so they would have seen something similar to this on YouTube and had somebody some people are applying these kinds of comments and that's how they learn and that's how they come up with the responses and the last example that I'm going to show to do with ytube is recommender systems and that's really what I work in is recommender systems and you'll know this from things like | 478 | 500 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=478s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | Netflix or Amazon but as well YouTube does recommend you videos and you homepage will be a recommended paint like a recommended set of videos that are specifically for you based on what you watched before and there's different ways that you can go by that so if you want to hear more about recommender systems that's my main research area that's the thing I know more about of | 500 | 523 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=500s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | like different ways that they can work so one way would be like people who like this video also liked this video and then another is you know finding similarities between the videos themselves and not actually taking into account the human side of things and then another thing would be taking into account people that are similar to you and finding what kind of videos they | 523 | 548 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=523s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | liked so um there's just a couple different ways if you want to hear more about recommender systems let me know in the comments below so now we're going to go into the types of learning algorithms very basic introduction to these different learning algorithms and just so that you have an understanding of the basic categories of machine learning so supervised learning and the first one | 548 | 570 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=548s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | we're going to talk about so the the main two categories are supervised and unsupervised learning but then there's also subcategories they're like semi-supervised and self supervised learning I'm gonna talk about those and then there's also reinforcement learning so I'm gonna talk about all of these in this video and but supervised learning basically what that means is you have a | 570 | 593 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=570s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | training data set with Associated labels so you essentially know the right answer in this situation and these labels have been provided by a human supervisor so somebody has labeled this data set so I've got the example that I was talking about here so we know the size of a house we know the amenities whether it's said facing your north's facing all of that kind of thing and then the | 593 | 616 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=593s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | accompanying housing price was obtained from whenever the house was sold so we have this training data which is actual houses and the actual housing prices they sell - and then you want to learn how to form the housing price based on the description so we have a description solution and we want to learn a general solution for when we have new descriptions but no solutions so for | 616 | 645 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=616s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | example we have a new house coming onto the market we know all of the information obviously we know that number of rooms number of bathrooms all of that stuff but obviously it hasn't gone to market we don't know the price but we can use the previous examples we can train a model on this data to kind of come up with an estimate of the housing price based on the fact that we | 645 | 668 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=645s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | don't know based on all the information we do know so that's the general supervised problem is we have training data with labels and we want to learn based on the training data features what would the label be and what issue with supervised learning always is that it's very hard to find label data and for bigger free for problems that have a ton of different options it can be hard | 668 | 696 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=668s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | to do so for example in speech recognition you know there's an endless number of combinations of words that could be said so to get somebody to come up with all of the possible sentences ever like that's not possible and then having labels so having the audio file and the accompanying labels it's just so difficult to come up with all of that so that's the problem with supervised | 696 | 721 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=696s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | learning is it's not always possible to get this kind of labeled data but then there are times that the housing prices that is definitely possible because you would have things like that on record and the good thing about modern times is that there's a ton of data being collected all the time by different apps and things like that and in those circumstances you often come up with a | 721 | 740 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=721s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | lot of labeled data naturally so unsupervised learning then is we have training data and it doesn't have labels so the goal here is to identify interesting structure in the data because we don't know the answer is necessarily but you can see in the graph they have associated here because I'm supervising something people don't grasp as easily because you're kind of thinking well what would | 740 | 764 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=740s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | be the point of this but one example is anomaly detection so that's when you're trying to something that's a complete outlier and it's either it's something new or it's just you know a total outlier so you can see we have in the data we have these clusters that we can see and obviously in a 2d data set like this it's very easy to see these clusters with your | 764 | 791 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=764s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | eyes not necessarily using any kind of algorithm so we can see we have these clusters of blue spots that are normal and then we have these orange ones that are being classified as noise because they don't fit into these clusters so those are the kinds of things you might want to look up for in a data set and obviously you're saying you know why do we need unsupervised algorithms if we | 791 | 812 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=791s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | can just see this but in a multi dimensional problem so here we've just got two variables if we have like 50 you can't see this with your own eyes when there's outliers like that but you can use an algorithm to cluster your data and then if we have data points that are so far away from the center of a cluster and I cluster is a group here so you can see them in the graph as well if we have | 812 | 837 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=812s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | something that's so far away from a cluster then we know it's probably an outlier or something like that so that's an example now we're gonna go into this sort of sub categories here of supervised and unsupervised so hopefully you have a general understanding of the difference here supervised we have labels unsupervised we don't and so the first example we're talking about is | 837 | 861 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=837s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | self supervised learning so we have an unlabeled data set but actually what people do in these self supervised learning is they actually turn an unsupervised they turn an unlabeled data set into a label data set by manipulating the data in some way so the basic task here is to find the missing part given and one part so to explain that better want to imagine if we have | 861 | 892 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=861s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | like what we had before was a set of sentences that was the data we were given what the um what the researcher does is they take as one part of and they generate this training data set with labels so we have the information is the trick the sentence with a piece missing and the answer is the piece that's missing so they train this way and this way we have labels because we have these | 892 | 924 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=892s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | sentence with the piece missing and we have the missing piece so they generate this label data set from unlabeled data so it works in these kinds of situations so in speech recognition well in sentence like if you're trying to understand the semantics of sentences you can take a sentence remove one piece and see if the system can learn to fill in the blanks and similarly with images | 924 | 948 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=924s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | we can take out a small part of the image and see kind of a system learn to spot fill in that missing part of the image and actually it has been relatively like pretty successfully be surprised if you want to look up self supervised learning it is pretty interesting to see how a system can learn these things and that's usually done with deep learning because it's not | 948 | 968 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=948s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | something a traditional machine learning algorithm can necessarily do myself like a linear model is not going to do that and so yeah basically in this sort of learning you want to learn about the underlying properties of the data and this would have required a lot more data if you don't do it this way so this is kind of like the system's being thrown into the deep end being expected to | 968 | 992 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=968s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | generate this missing piece from basic examples that they've seen before so moving on we have semi-supervised learning so I mentioned before that labels are very hard to obtain and especially in cases like speech recognition and web content classification so if you've got a webpage and you want to add tags to it you know it's hard to it's very time consuming for a person to sit down and | 992 | 1,023 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=992s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | read through that entire webpage and manually assign tags and then as well you probably have to have a couple of different people doing it so that they can it's not like it's one person's bias it's a few people who are collectively doing so it's very time-consuming it takes a lot of Human Resources to do all of that labeling so semi-supervised learning algorithms have been developed for the | 1,023 | 1,046 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1023s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | cases where we only have a small amount of labeled data and I'm a very small amount when there is a huge amount of data that could potentially be there so what I said about sentences there's unlimited possibilities for sentences and we might have a small number of labeled audio to sentence and files and a ton of audio data that haven't been labeled so basically this is used in any | 1,046 | 1,071 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1046s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | task that requires a big amount of human resources to do the labeling and when there's a big amount of possibilities so the characteristics here are that we have a huge amount of unlabeled data otherwise we would just use a supervised learning algorithm and there's an input/output proximity symmetry so what does that mean it basically means that the underlying requirement here is that | 1,071 | 1,092 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1071s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | if two inputs are similar their output should be similar and that's a very simple thing to understand but it's if it's not present this won't work because you can't make things you can't get the system to learn that to some inputs that are similar should be very different you know this kind of algorithm wouldn't make sense in that way so that's one other requirement and another | 1,092 | 1,117 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1092s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | requirement is that the labeling should be relatively simple so we have we need we need the simpler labeling to be not more difficult if we have this middle step so trying to get the labels from the small set of label data should not be more difficult than it would be if we had humans doing it as well like generally we'd have this is a low dimension problem meaning that there's | 1,117 | 1,141 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1117s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | not a ton of features and a ton of input variables and things like that so the last one we're going to talk about is reinforcement learning and this definitely could do with this whole other a whole other video all about reinforcement learning but I just wanted to give a basic introduction to it because I think that's something people are usually trying to figure out what's | 1,141 | 1,160 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1141s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | the difference between supervisor unsupervised and reinforcement learning but basically this is a form of supervised learning the the main factor here is the system focuses on maximizing a reward signal so the reward signal is supervision that that is a convenient so we have an agent and it's said that into the feature space so all of the input variable options are there and it was it has to | 1,160 | 1,190 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1160s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | learn what actions it can take in this feature space in order to maximize the reward signal so one example would be AI chess or any form of AI game playing it's generally done with reinforcement learning so it's if you example if you think about the chess board the the thought I guess that's playing chess it has all of these options of the different moves they can take and the | 1,190 | 1,217 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1190s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | reward they're trying to maximize is winning the game and that's all that really matters and this is whether they win the game or not so it'll play this game hundreds and hundreds of times until it learns what exact moves in both scenarios so that the environment also plays a role here so you can see in the graph we've got the agent actions environment and so here the agent would | 1,217 | 1,243 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1217s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | be the ball playing chess the action will be a move that the bot takes the environment is in this case we have somebody else playing chess so the move that they take is the sort of environment that they need to relate to as well as the chess board that results from these actions so when the bot makes a move their pieces move so they need to update their understanding of the moves | 1,243 | 1,272 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1243s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | that are available based on the move that they've taken and then again by the moves that the other person has taken and then so the state will be the game the board game the layout of the board essentially and the reward again will just be the whether they're winning or not their probability of winning will probably be included in the some way based on the layer of the board | 1,272 | 1,298 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1272s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | and they learned by trial and error what is the best move to make in what situation and there's also usually a delayed rewards obviously in this case we have that there is the winning of the game and that can be delayed in the sense that some moves that you take will have a knock-on effect on your probability of winning so like if this game is tracking the probability of | 1,298 | 1,324 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1298s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | winning your next move might not effect that necessarily but you'll give you access to the move after and that could increase by a lot another example is things like a but that has to travel around and collect things obviously every move that it takes won't necessarily collect something but a few moves from there if we continue along the right track they'll pick up something and that will | 1,324 | 1,348 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1324s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | be a reward so the delayed ward is a big characteristic there and then because that's how it learns how this set of moves are important as not just a single move but like a series of moves and then the trial by error is how it learns and yeah usually there'll be like hundreds and hundreds and thousands of instances of this and then that's how they learn so again reinforcement learning can do | 1,348 | 1,373 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1348s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | its whole other video I just wanted to give a basic introduction to all of these things so I really hope that that was helpful for you guys I hope you enjoyed this video and if you want to see more videos like this again do be sure to give it a thumbs up onto um comment down below because this is a new style of video on this channel and when we're trying get new things here I need | 1,373 | 1,395 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1373s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
2Z1B0xESzMw | to know they're working so buying you guys giving the video a thumbs up on commenting below it then that is way for me to understand whether it's working or not and then in future that means I can sort of delve into the individual algorithms that you'll see a machine learning and probably I'll do one week on the actual algorithm and explaining that and then the following week on the | 1,395 | 1,416 | https://www.youtube.com/watch?v=2Z1B0xESzMw&t=1395s | Supervised vs Unsupervised vs Semi / Self Supervised vs Reinforcement Learning | Machine Learning | |
Uumd2zOOz60 | hi there people so a lot of you have asked me how I read papers and honestly I don't think there is any super special method to it but you know I thought because people have asked me to make a video on it so I'll make a video on it and I'll try to share my method of reading papers and hopefully this is going to be somewhat of a miniseries or a series where I every now and then | 0 | 28 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=0s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | discuss how I read one of the papers that I make videos about and I'll try to select them such that different things are highlighted now I've selected this one right here for really for no particular reason other than I sort of remembered it and going to try to go with you through how I read this and how I encountered this and kind of try to honestly share what I thought at the | 28 | 56 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=28s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | first time when I read it and I hope this helps some of you if it does help you and if you like content like this of course feel free to share this out and subscribe if you haven't seen my original video on this paper it might be worth to go watch it I'll link it and with that let's dive in so again this might be not really something new but I'll just go through it okay so first | 56 | 87 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=56s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | thing I do is of course read the title so the title has three parts and to end object detection with transformers so what I notice that I do myself is I like through reading a paper it's like read the paper with an open mind I don't do that I almost immediately form an opinion and a hypothesis of what's going on like so I see transformers so I know what transformers are if you don't I've made | 87 | 114 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=87s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | a video if made lots of videos on transformers attention is all you need is the base paper for that so I know what a transformer is okay and I know that transformers are usually in NLP are usually used in NLP door there are things like you know other things with transformers but usually an NLP model then I read object detection and I know object detection is | 114 | 139 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=114s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | a computer vision tasks so immediately this here is sort of a a difference and I immediately try to assess what's the new idea in this paper and in this case it might be okay it might be applying transformers to object detection but then I also see end to end and the only reason to put that in a title is because that's the novelty because usually in deep learning we're sort of used to | 139 | 163 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=139s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | systems being and to end and even if they aren't if most systems aren't end-to-end a lot of people don't is like end to end image classification on image net like Thanks so I I was guessing that the reason they put in end-to-end into the title was because that's actually something that's special about the model so now I have like two competing hypotheses of why this paper matters | 163 | 190 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=163s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | first of all because it does it with transformers and second because it does it end to end and the of course the true fear is that the combination of end to end transformers all of that is what makes this model and I already form like a hypothesis of whether I like this or not like I have to be have to be honest I have very quick judgment of papers of whether I like them or not and then I | 190 | 216 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=190s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | sort of catch myself each time and I still try to so they're for most papers actually that I have sort of a negative opinion at the beginning where i will- there are papers where I think like there is no way this is going to you know work or something like this I'm actually positively convinced throughout the paper so for most for most papers where I that I read I'm trying to find | 216 | 246 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=216s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | the positive things in there but I do form an opinion pretty quickly usually alright so the second thing this part right here I like I don't even see this is like advertisements on on like Twitter I like you you just I have always had issues with author names like people would come to me and be like oh have you seen the new vineos paper and no clue and then when they say like oh | 246 | 273 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=246s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | that's where they use this character level model to do that and I'm like oh that paper so I like do not care who the authors are of a paper like I don't I can't remember the papers by their author names I've gotten better at it I have to say but i-i've always had trouble with this now that's not to say that a name doesn't pop out to me like if this would be like it like yoshua | 273 | 297 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=273s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | bengio or some someone like really famous then of course that would catch my eye and but I also know that you know yoshua bengio is paper yoshua bengio slab is huge so just because a big name is on the paper doesn't mean that the paper is going to be of any good or bad quality sometimes the author's give you an indication of what kind of research is there like if | 297 | 322 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=297s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | you see a Najaf Clun or Ken Kenneth Oh Stanley you know that there's there's going to be the certain type of of you know learning to explore and and and kind of a bit more out-of-the-box thinking in their papers which I really like but it doesn't immediately give you a clue maybe if you go by first authors it's much more indicative if you have already read some of their papers but | 322 | 351 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=322s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | most often I just ignore authors and go on the affiliation sometimes matters in in that it's a bit of a vicious cycle if there's a big name affiliation like Facebook AI Google ai and so on these papers also they get more exposure in like the press and so on so whatever Google a publish that paper all of these all these pops eye magazines like verge and this and life hacker and hacker news | 351 | 380 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=351s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | and whatnot they like like write a blurb about it so often they get much more scrutinized for these papers they get much more they get much more the public attention but they also get much more scrutiny which in turn means that there is a bit more pressure on them to do good experiments so that biases meet like a little bit into the direction of believing their experimental evidence | 380 | 408 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=380s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | more now usually this is also backed up by the fact that I am actually convinced by their experiments usually so these these big-name papers often I find myself even without or disregarding the affiliation to be convinced more than of like regular papers my most often issue with papers is that I don't believe the experiments and I make no difference like even if | 408 | 437 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=408s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | it's Facebook I still my prior is the experiments or crap and I don't believe them and they have to convince me of the opposite but some like I can't say that it doesn't affect me that it's like a big name affiliation okay so then the second thing is I sometimes I see the paper on archive and I skim the abstract sometimes the abstract is informative and sometimes not so here it's like blah | 437 | 467 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=437s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | blah blah a new method that views object detection as a direct set prediction problem I'm like oh yeah okay so streamlines the detection effectively removing the need for many hand design components like non maximum suppression yada yada yada the main ingredients called detection transformer asset based global loss that forces unique prediction via bipartite matching and | 467 | 490 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=467s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | the transformer encode or decode or architecture so they make it clear here why it matters and that's that's what I what I want to get at is sort of what's the new thing in this paper most papers are even though they're all very long and have lots of math and so on they often have like one or maybe two new core things that they really tell you sometimes zero but a lot of times | 490 | 517 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=490s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | it's like one thing that they really do and you you sort of have to but they're trying to cloak it often because they need to make their research as impactful as possible right but you need to sort of figure out what it is they're doing they make it fairly easy for us in that they say okay they remove the need for many hand designed components like non maximum suppression which tells me that | 517 | 543 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=517s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | they're building something that's easier than what came before them and that already tells me it it's not necessarily going to be better their argument is more that it's going to be easier right there there are sort of two kinds of experimental results the ones where you try to beat what came before you and the ones where you're trying to say look our thing works just as well as this other | 543 | 567 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=543s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | thing while being more advantageous in some other metric so I would place this already in the sort of second category and then they say what are the actual ingredients it's a set based global loss that forces unique predictions via bipartite matching now I at this point I know what these terms mean but at this point actually don't have to know what the terms mean what I need to recognize | 567 | 591 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=567s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | is that I simply have to go later and figure out what that is and a transformer based encoder/decoder architecture okay so there are two things right here that I remember I need to pay attention to later there is this loss which seems to be special and there is the transformer architecture which seemed which they say okay that that's the model basically consists of those | 591 | 618 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=591s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | two things and then they have a short description of what it does given a fixed small set of learned object queries that your reasons about the relations of the objects and the global image context to directly output the final set of predicted in parallel that almost tells me nothing of yeah okay the model reasons maybe this in parallel is something but the model is conceptually | 618 | 645 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=618s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | simple and does not require specialized library unlike many other modern detectors this sort of repeats this enforces my hypothesis that they're going with the hey this is a much easier way of doing things approach detter demonstrates accuracy and runtime performance on par with well established that further confirms my hypothesis that this is on are right they runtime performance on | 645 | 670 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=645s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | par with the current state of the art and at the end they say moreover data can easily be generalized to proceed to produce Panoptix segmentation in a unified manner we show that it significantly outperforms competitive baselines training code and praetor models are available as part when I first read it is like ok can easily be generalized to produce this Panoptix | 670 | 694 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=670s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | segmentation this is I didn't know yet whether this is like a central claim of their paper that it can do this segmentation or whether this is like an added benefit to their paper because you can read it in both ways and I'm just ready to find this out in the paper now after I've read the abstract and sort of already form the hypothesis of what's going on so here I already in my mind I | 694 | 718 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=694s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | already sort of have a model of how would I do that right how would I how would I do that and then what would I do so right now what I might be thinking is if I have a transformer over images that directly outputs the the predictions in parallel I'm imagining like an image and the image somehow needs to go into a transformer so maybe there's like an encoder like a CNN encoder that gives me | 718 | 750 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=718s | How I Read a Paper: Facebook's DETR (Video Tutorial) | |
Uumd2zOOz60 | image features and then it's it's so maybe you sample this down this image this is just me hypothesizing what could be going on right and then I might be unrolling that right this image into a vector of these lower pixels and then so in my mind what I would do right here if without knowing anything more would be to do something like Bert's pan predictions so I would have Bert right | 750 | 779 | https://www.youtube.com/watch?v=Uumd2zOOz60&t=750s | How I Read a Paper: Facebook's DETR (Video Tutorial) |