So. Unsupervised Learning of the 1980s is NOT unsupervised learning of today. Actually it should be called predictive learning but most people don't bother the distinction. Thanks for the clarification!
I think people learn by a linear combination of all three categories supervised, reinforcement and unsupervised learning. We are told how the world works, we are told what we did right and wrong and we can without anyone's input perceive associations between inputs and outputs. We seem to be able to combine all these.
we aren't told how the world works. A teacher or your parents telling you something is just more noise patterns which could be dealt with under unsupervised learning. Our equivalent of supervised learning would have to be some perfect voice of god telling us perfect truths. No one ever gets perfect truths like that. Your teacher could be lying to you, they could be wrong, and even the language you are using has inconsistencies and contradictions. It is all unsupervised your whole life. You build up a general correlational understanding all from unsupervised learning.
anytime another person shows you examples of an action and the outcome of that action that would be supervised learning. having the person repeat themselves or them beating you and yelling at you bad ausroy would be negative reinforcement learning, and having no other person involved is unsupervised
No it's not supervised. Again, supervised is where there is a perfect answer that is "main-lined" into the system. You seem to attach some special significance to 'another person' showing you something, as if they aren't fallible, a single observation and just part of the whole ecosystem of causes. Someone showing you an action and an outcome is just pure unsupervised observation, like you would watch a rock fall, like you would watch a tree catch on fire. You can ignore them, you can doubt their knowledge, the same action may have a different outcome the next time. There is no main-lining perfect knowledge like supervised machine learning is. The mind operates in the same way whether you are watching a rock fall or having a person show you something. There's no magic about another person being there that makes it supervised. Supervised machine learning is abnormal magic learning though.
it seems like you are trying prove a contradiction to reinforce the idea that the only way I can learn something is through my own unsupervised trial and error.
need help with one confusion. LeCun talks about using Mask R-CNN to detect distinct objects and he says that this is cutting edge. i see the video was recorded on Nov 1st 2017. How is this technology superior to what my phone camera does when it detects faces and guesses (incorrectly) ages of people in the pic? What is the technological leap between the two?
Good speech. But the most annoying part of this video is switching between slides and the speaker. I don't want/need to see his face all the time. I want to see the slides but slides are swapped so quickly and I don't have time to go though the slide which has a lot of useful information. Eventually I took screenshot of the slide and put it on my screen so I can look at it when he is talking about it.
I don't think it is possible to relay all the important technical information while still keeping the talks accessible for the general population. I believe what is important is to be able to fully explain and go through the technical details so that other developers can get a better understanding and help the field progress faster. To relay the information to a non technical audience it would most likely be better to have a different talk altogether that is tailored specifically to that group.
I don't think your iq is too low, i just think your interests are misplaced (you should probably stop watching videos about male porn stars and learn to master one field you're interested in before moving on to the next one for context)
One of the most accessible and informative videos I’ve seen on deep learning. Thanks for sharing this!
All I heard was "things got fast enough so we can finally try out deep networks models".
He does not say how deep learning actually works. He just says what technologies are avaiable today.
So. Unsupervised Learning of the 1980s is NOT unsupervised learning of today. Actually it should be called predictive learning but most people don't bother the distinction. Thanks for the clarification!
Some recommenders systems are unsupervised but can make predictions
Very recommendable! LeCun can be at times very funny too! Good overview of state of the art!
Amazing Lecture. Thank you for the upload.
Excellent video and lecture!
I think people learn by a linear combination of all three categories supervised, reinforcement and unsupervised learning. We are told how the world works, we are told what we did right and wrong and we can without anyone's input perceive associations between inputs and outputs. We seem to be able to combine all these.
we aren't told how the world works. A teacher or your parents telling you something is just more noise patterns which could be dealt with under unsupervised learning. Our equivalent of supervised learning would have to be some perfect voice of god telling us perfect truths. No one ever gets perfect truths like that. Your teacher could be lying to you, they could be wrong, and even the language you are using has inconsistencies and contradictions. It is all unsupervised your whole life. You build up a general correlational understanding all from unsupervised learning.
anytime another person shows you examples of an action and the outcome of that action that would be supervised learning. having the person repeat themselves or them beating you and yelling at you bad ausroy would be negative reinforcement learning, and having no other person involved is unsupervised
No it's not supervised. Again, supervised is where there is a perfect answer that is "main-lined" into the system. You seem to attach some special significance to 'another person' showing you something, as if they aren't fallible, a single observation and just part of the whole ecosystem of causes. Someone showing you an action and an outcome is just pure unsupervised observation, like you would watch a rock fall, like you would watch a tree catch on fire. You can ignore them, you can doubt their knowledge, the same action may have a different outcome the next time. There is no main-lining perfect knowledge like supervised machine learning is. The mind operates in the same way whether you are watching a rock fall or having a person show you something. There's no magic about another person being there that makes it supervised. Supervised machine learning is abnormal magic learning though.
it seems like you are trying prove a contradiction to reinforce the idea that the only way I can learn something is through my own unsupervised trial and error.
What contradiction? And yes, I am trying to reinforce the idea that human learning is unsupervised. That seems useless to say.
need help with one confusion. LeCun talks about using Mask R-CNN to detect distinct objects and he says that this is cutting edge. i see the video was recorded on Nov 1st 2017. How is this technology superior to what my phone camera does when it detects faces and guesses (incorrectly) ages of people in the pic? What is the technological leap between the two?
image segmentation is much harder than image classification
thank you for this vid
Okey great! But I don't get any of that
12:00 terminator scanning is coming.
Good speech. But the most annoying part of this video is switching between slides and the speaker. I don't want/need to see his face all the time. I want to see the slides but slides are swapped so quickly and I don't have time to go though the slide which has a lot of useful information. Eventually I took screenshot of the slide and put it on my screen so I can look at it when he is talking about it.
I'm from four years in the future, and I've come to show you the magic of the pause button...
I work with your nephew and he loves JavaScript looool
only technical people can enjoy his presentation! Not good for general audience
I don't think it is possible to relay all the important technical information while still keeping the talks accessible for the general population. I believe what is important is to be able to fully explain and go through the technical details so that other developers can get a better understanding and help the field progress faster. To relay the information to a non technical audience it would most likely be better to have a different talk altogether that is tailored specifically to that group.
Build an AI that predicts the birthdate of the singularity.
Hey, let's build an AI that is able to give presentations about AI that won't put humans to sleep, that possible???
hahah wow, that's sad. i hope you find a life soon!
I have two good news for you: (1) there are a lot of cat videos in the internet, and (2) there are AI systems that correct grammar.
you should not watch those subjects, because ur IQ is too low to understand anything about AI. what a pathetic crown.
I don't think your iq is too low, i just think your interests are misplaced (you should probably stop watching videos about male porn stars and learn to master one field you're interested in before moving on to the next one for context)
Switch this blasphemous abomination off