I have no idea where you are from, but I have studied in two continents, 3 different universities, and this was my experience in all of o them. Academia is just an amazing world.
@@ev.c6until some people try to get popular by changing the data and embellishing things. Bad apples yes, but they look the most appetizing until you bite into one.
Read more about Alexei Efros's research in a written interview by Susan D'Agostino on the Quanta website: www.quantamagazine.org/the-computing-pioneer-helping-ai-see-20231024/ Quanta is conducting a series of surveys to better serve our audience. Take our video audience survey and you will be entered to win free Quanta merchandise: quantamag.typeform.com/video
I am waiting for a video on the progress of Quantum Optics. 😃 I am hoping to pursue research in this field and it has some of the greatest ideas of all of experimental physics.
Made me smile in the same way. One of the first things my professor told me at the beginning of the phd was that his goal is to make me a better scientist than him. Really nice moment to see this guy so passionate about it as well.
AI generated timestamps 0:00: 👁 Computer vision is a complex process that is difficult for computers to replicate, but advancements are being made. 2:56: 🌳 Visual data and its importance in machine learning and computer vision. 5:58: 🔑 Computers struggle to generalize in their machine learning algorithms, but test time training can help improve their performance.
Computer vision is hard because the base systems of encoding do not contain discrete elements like biology. Which means that computer vision with or without AI is unable to generate discrete representations of elements in the real world. And that applies to objects, textures, colors, lights and so forth, all of which are what makes up the "real world". Biological vision doesn't need labels in order to recognize shapes and patterns because those elements are formed using discrete biological encoding of light information, which makes it easier to distinguish one color, shape, texture and object from another, ie discrete elements in 3d space.
All very interesting. I wonder if we are limiting computer vision by only considering human vision. Each other organism has vision selected to make the organism successful, and its not like ours. I wonder if there is something we can learn from this diversity of purpose for visual systems in all organisms. Alexei Efros has touched on this diversity of purpose with his own experience of vision.
yeah well computer vision in ranges of the electromagnetic spectrum outside of visible light exist. That is more relevant to hardware: how the sensor is detecting light and what range of frequencies etc. Once it becomes image data of whatever kind, the convolutional neural networks do their thing and don't really care about how "humans" see things.
@@dexterrity There also sonar for bats and other creatures, but I was thinking more about the cognitive processes, although yes, the hardware is certainly required.
@@TzaraDuchamp Efros made a point of his personal experience with low vision which helped him move forward. I was just proposing that perhaps we could move forward by considering a broader specturm of experience by tapping into animal vision. Its not about how computers currently perform computer vision algorithms, its about learning how we could uncover insights that allows us to enhance or redesign computer vision.
@@TzaraDuchamp You misunderstood me - I was wondering if we could get more insight from a broader view. I didn't cast any aspersions on Efros - in fact I admire the man. Maybe reading too much between the lines?
This is a very good interview. I am glad to see that it's validating my intuition, about the fact that models should continuously learn instead to being frozen, and then retrained from scratch. One of the biggest difficulties to improve the current techniques is reducing models size. I don't know how much data a real brain can store, but given the miniaturization of current chips, I suspect we are wasting resouces. Anecdote: I have bad eyesight as well. 😂
Computer scientist Alexei Efros suffers from poor eyesight, but this has hardly been a professional setback. It's helped him understand how computers can learn to see. At the Berkeley Artificial Intelligence Research Lab, Efros combines massive online data sets with machine learning algorithms to understand, model and re-create the visual world. His work is used in iPhones, Adobe Photoshop, self-driving car technology, and robotics. In 2016, the Association for Computing Machinery awarded him its Prize in Computing for his work creating realistic synthetic images, calling him an “image alchemist.” In this video, Efros talks about the challenges and changing paradigms of computer vision.
I had an idea when I was working on my thesis that if we have transformer for vision and a new embedding system that treat the visual data like human we can have a model that will understand the images of the universe that is beyond the computer ability of human brains such as the cosmic microwave background. But it’s an idea only😢
Interesting to see the distribution of ethnicities along that outside shot bench.. humans are drawn to those with whom they assume they might have common ground. Just an observation. Might be wrong.
In all but simple artificial environments, picking out what is significant in a view is not logically analysable. It is not logical but axiological. It depends what a subject notices, or is looking for. Manufacturing yes, where the choices have been artificially limited, basic cognition no, where it is unclear what we are even looking for, if anything.
Basics are a bit hard to understand, lot less opportunities, only big tech is working on computer vision right now, which only select algorithmically strong candidates who have been vetted from multiple selection programs, be it college entrace tests, then GPA, then previous internships (starting is the biggest hurdle).... After all this, we get to taste the basics of Computer Vision in bigtech, to understand basics one needs at least 3years of dedicated grasping + implementation... till then the world has moved on to something more advanced
the problem is that even if you watch a real video from nature on the screen, it is not real for your eyes, a two-dimensional image plus unrealistic colors of the screen, i.e. resolution..
The next big step will be generalisation. For example, when AI can infer from its training data generally what a road looks like whether it has snow, leaves or sand on it.
We literally have cameras for a few centuries now, making AI learn to "see" is just that, a camera attached to AI processing it, we already feed AI with pics and make it learn visually
There are multiple levels of vision. Everything from pattern matching is to recognizing symbols to identifying and interacting with objects. We see mostly with our brains, for instance.
@@JuliusUniquewell usually you train a model on the dataset of images or videos then once it is trained you can test its capabilities by feeding an input image/video that wasnt in the training data now this is just a very simplified explanation and its more complex than that
Still not "AI" and this exploitation of the term is exhausting. He even admits its about data comprehension ie algorithmic formulations (tiered) and not unprovoked generation which is and was the metric for the term. We have lost the boundaries of what things are so as to cater to branding for $$$
its exactly AI, what are you talking about? maybe very old Computer vision was, recent research into the domain is all AI. If anything, Computer vision was the field impacted most by AIl, especially in early days of deep learning.
@@khalilsabri7978 You could then assign any and every computational process as "AI" based on the metrics you and they are suggesting wildly. What was once labeled "bots" with keyword association generative replies are now "AI" bcz every thing has been rebranded to serve a new narrative for profit. AI used to have a requisite to meet in order to be classified as AI, we had science fiction esk tests as thresholds, and if you can claim any of these things just abundantly appearing all of a sudden today meet those standards, then you are a mindless consumer. Image generation from keywords is not AI its is algorithmic compiling. ChatGPT is just search aggregation with a fancy front end. None of these things generate information independent of the user defined rules or software defined boundaries, thus why it is so easy to censor information immediately. As for research, literally nothing has changed.. data is compiled, an algorithmic is authored to seek a model, where is the AI?
Unprovoked generation is and was the metric for the term in which field? Computer science, or science fiction and general aspiration? Thinking of early intelligence in single-celled life, a part of it must have been in reacting to light when moving around in the water. Seeing energy, food, and the environment. Is that not intelligence enough for something not alive yet to be able to autonomously sense and react to the world. Artificial intelligence for me should connect all modes of sensing and making inferences into a single place. Then, computer vision is exactly AI in the same sense as computer generation "unprovoked" or not.
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
I love that with 120.000 citations, he is regarding the grad students and the next generation of scientists as his biggest achievement.
It's great that there are professors out there that value their students as their greatest achievement!
I have no idea where you are from, but I have studied in two continents, 3 different universities, and this was my experience in all of o them. Academia is just an amazing world.
@@ev.c6then u r lucky that you got this kind of experience bcz mine wasn't.😅
@@ev.c6until some people try to get popular by changing the data and embellishing things. Bad apples yes, but they look the most appetizing until you bite into one.
Hiw do they work so hard for so long and not get bored and tired and frustrated?
@@blueAndblack-ec6jkis working 8 hours a day enough as a grad student so it doesn't have to fucking wear you out or take over your life?
Read more about Alexei Efros's research in a written interview by Susan D'Agostino on the Quanta website: www.quantamagazine.org/the-computing-pioneer-helping-ai-see-20231024/
Quanta is conducting a series of surveys to better serve our audience. Take our video audience survey and you will be entered to win free Quanta merchandise: quantamag.typeform.com/video
I am waiting for a video on the progress of Quantum Optics. 😃 I am hoping to pursue research in this field and it has some of the greatest ideas of all of experimental physics.
As a computer scientist working in Computer Vision tasks (and other AI applications) for medical imaging processing, this video made me smile :)
In a good way?
Made me smile in the same way. One of the first things my professor told me at the beginning of the phd was that his goal is to make me a better scientist than him. Really nice moment to see this guy so passionate about it as well.
As some random guy sick of seeing these subtle humble brag comments, your comment made me cringe
Next time I’ll be more modest @nutmeg0144 :)
All they said was that they work in the field and enjoyed seeing the video? The only thing cringe was your response@@nutmeg0144
I love how at 8:08 one of the students' phone falls out of their pocket and everyone turns and looks at it
AI generated timestamps
0:00: 👁 Computer vision is a complex process that is difficult for computers to replicate, but advancements are being made.
2:56: 🌳 Visual data and its importance in machine learning and computer vision.
5:58: 🔑 Computers struggle to generalize in their machine learning algorithms, but test time training can help improve their performance.
wow
Wow
Were the emojis from the AI too?
yup @@mihailmilev9909
Wonderful video! I love everything this channel has made!
my favorite topic in CS
At minute 6:26, the scene is Havana, Cuba. Near where I live🙌
Computer vision is hard because the base systems of encoding do not contain discrete elements like biology. Which means that computer vision with or without AI is unable to generate discrete representations of elements in the real world. And that applies to objects, textures, colors, lights and so forth, all of which are what makes up the "real world". Biological vision doesn't need labels in order to recognize shapes and patterns because those elements are formed using discrete biological encoding of light information, which makes it easier to distinguish one color, shape, texture and object from another, ie discrete elements in 3d space.
All very interesting. I wonder if we are limiting computer vision by only considering human vision. Each other organism has vision selected to make the organism successful, and its not like ours. I wonder if there is something we can learn from this diversity of purpose for visual systems in all organisms. Alexei Efros has touched on this diversity of purpose with his own experience of vision.
yeah well computer vision in ranges of the electromagnetic spectrum outside of visible light exist. That is more relevant to hardware: how the sensor is detecting light and what range of frequencies etc. Once it becomes image data of whatever kind, the convolutional neural networks do their thing and don't really care about how "humans" see things.
@@dexterrity There also sonar for bats and other creatures, but I was thinking more about the cognitive processes, although yes, the hardware is certainly required.
@@TzaraDuchamp Efros made a point of his personal experience with low vision which helped him move forward. I was just proposing that perhaps we could move forward by considering a broader specturm of experience by tapping into animal vision. Its not about how computers currently perform computer vision algorithms, its about learning how we could uncover insights that allows us to enhance or redesign computer vision.
First problem is that humans are creating AI. We are going to be AI's limit
@@TzaraDuchamp You misunderstood me - I was wondering if we could get more insight from a broader view. I didn't cast any aspersions on Efros - in fact I admire the man. Maybe reading too much between the lines?
Love the short video!❤
This is a very good interview. I am glad to see that it's validating my intuition, about the fact that models should continuously learn instead to being frozen, and then retrained from scratch.
One of the biggest difficulties to improve the current techniques is reducing models size. I don't know how much data a real brain can store, but given the miniaturization of current chips, I suspect we are wasting resouces.
Anecdote: I have bad eyesight as well. 😂
Thank you for the insights and this very well produced video!
Computer scientist Alexei Efros suffers from poor eyesight, but this has hardly been a professional setback. It's helped him understand how computers can learn to see.
At the Berkeley Artificial Intelligence Research Lab, Efros combines massive online data sets with machine learning algorithms to understand, model and re-create the visual world. His work is used in iPhones, Adobe Photoshop, self-driving car technology, and robotics. In 2016, the Association for Computing Machinery awarded him its Prize in Computing for his work creating realistic synthetic images, calling him an “image alchemist.”
In this video, Efros talks about the challenges and changing paradigms of computer vision.
Wonderful! Looking forward to the future!
Man.. I wish you were my CS professor. 👍
I had an idea when I was working on my thesis that if we have transformer for vision and a new embedding system that treat the visual data like human we can have a model that will understand the images of the universe that is beyond the computer ability of human brains such as the cosmic microwave background. But it’s an idea only😢
Love this channel
so amazing.😍😍🤩🤩.good luck.
thank you for explanation!
1:55 Im not sure if showing birds was intentional here. In any case, it looks like a nod Gregor Mendel's genetics work.
Nice informative video.
Thank you👍
Interesting to see the distribution of ethnicities along that outside shot bench.. humans are drawn to those with whom they assume they might have common ground. Just an observation. Might be wrong.
5:28 he is so deep inside, he calls us 'agents'
Computer vision is so fun!
In all but simple artificial environments, picking out what is significant in a view is not logically analysable. It is not logical but axiological. It depends what a subject notices, or is looking for. Manufacturing yes, where the choices have been artificially limited, basic cognition no, where it is unclear what we are even looking for, if anything.
Amazing!
what about use analogue computing in the futur for AI ?
Cool!!!❤❤
Computers cannot see, and will never see, they only process information, but will never see.
Thumbnail lookin’ like a front foot catch 3 flip
I also have Myopia
Basics are a bit hard to understand, lot less opportunities, only big tech is working on computer vision right now, which only select algorithmically strong candidates who have been vetted from multiple selection programs, be it college entrace tests, then GPA, then previous internships (starting is the biggest hurdle)....
After all this, we get to taste the basics of Computer Vision in bigtech, to understand basics one needs at least 3years of dedicated grasping + implementation... till then the world has moved on to something more advanced
What about computer audition?
Waiting for the day when computer vision beat skills of georainbolt
the problem is that even if you watch a real video from nature on the screen, it is not real for your eyes, a two-dimensional image plus unrealistic colors of the screen, i.e. resolution..
Computer vision is hard because it's right at the mercy of the so-called curse of dimensionality.
thx for supporting Ukraine
The next big step will be generalisation. For example, when AI can infer from its training data generally what a road looks like whether it has snow, leaves or sand on it.
So AI is just data with some selective results from that data ..is it ?
the best question is how tesla computer vision works
"UNO CHE R K".Geo mittchelin
3:35 Slava Ukraini
matlab !!
Two minute paper 😊
We literally have cameras for a few centuries now, making AI learn to "see" is just that, a camera attached to AI processing it, we already feed AI with pics and make it learn visually
There are multiple levels of vision. Everything from pattern matching is to recognizing symbols to identifying and interacting with objects. We see mostly with our brains, for instance.
@@jsmunroe I thought it's just having a lot of digital neurons and then letting them figure out the concept of patterns themselves
@@JuliusUniquewell usually you train a model on the dataset of images or videos
then once it is trained you can test its capabilities by feeding an input image/video that wasnt in the training data
now this is just a very simplified explanation and its more complex than that
you didn't explain how AI learns to see, like at all, i'm gonna have to give a thumbs down
Panoptic segmentation is to complicated for an eight minute video
cool and first comment
I was early.
ёр инглиш из вери велл
Just convert a 2d plane to 3d calculations 😂
that's how our brain works converting 3D into 2D then analysing the image
Geo NRA 16 000 000 000 000 000 BANK.
Still not "AI" and this exploitation of the term is exhausting. He even admits its about data comprehension ie algorithmic formulations (tiered) and not unprovoked generation which is and was the metric for the term. We have lost the boundaries of what things are so as to cater to branding for $$$
yes hype and money!!!
its exactly AI, what are you talking about? maybe very old Computer vision was, recent research into the domain is all AI. If anything, Computer vision was the field impacted most by AIl, especially in early days of deep learning.
@@khalilsabri7978 You could then assign any and every computational process as "AI" based on the metrics you and they are suggesting wildly. What was once labeled "bots" with keyword association generative replies are now "AI" bcz every thing has been rebranded to serve a new narrative for profit. AI used to have a requisite to meet in order to be classified as AI, we had science fiction esk tests as thresholds, and if you can claim any of these things just abundantly appearing all of a sudden today meet those standards, then you are a mindless consumer. Image generation from keywords is not AI its is algorithmic compiling. ChatGPT is just search aggregation with a fancy front end. None of these things generate information independent of the user defined rules or software defined boundaries, thus why it is so easy to censor information immediately. As for research, literally nothing has changed.. data is compiled, an algorithmic is authored to seek a model, where is the AI?
Unprovoked generation is and was the metric for the term in which field? Computer science, or science fiction and general aspiration?
Thinking of early intelligence in single-celled life, a part of it must have been in reacting to light when moving around in the water. Seeing energy, food, and the environment. Is that not intelligence enough for something not alive yet to be able to autonomously sense and react to the world.
Artificial intelligence for me should connect all modes of sensing and making inferences into a single place. Then, computer vision is exactly AI in the same sense as computer generation "unprovoked" or not.
Vision is hard problem for.humans and animals too. We need a lot of frames and points of view to figure things out, and still make a lot of mistakes.
GRCE 1314 .Geo