|
Teal posted:The critical difference between human memory and an autoencoder is that we build the representation on basis of abstract yet meaningful parallels and there's no proof this will ever work by feeding a "large enough" neural net based autoencoder of this type "enough data". It don't seems to be a huge difference. Or maybe I am too dumb to understand any of this. It don't sounds like autoencoders work very differently than human memory, only that human memory uses another subsystem (a tokenizer/ a lexer phase) that autoencoders don't include.
|
#
?
Dec 27, 2017 17:02
|
|
|
|
| # ? May 8, 2024 13:29 |
|
|
Tei posted:It don't seems to be a huge difference. Or maybe I am too dumb to understand any of this. It would be easier to formulate the differences in exact and specific terms if we had complete understanding of how our mind and memory works. For one thing; what the other guy said - one of the biggest hurdles of Deep Learning development that a lot of people is trying to figure a way around is that NNs are generally very spatially sensitive and for some specific input space (for instance XY raster image), if you want to distinguish an apple no matter where it is in the picture, you ideally need to show them a more or less uniform distribution of apples in all possible positions in the image - if all your samples had an apple in the bottom right corner, you'd hit a wonderful accuracy and all, but a new sample with an apple in top left corner would get misclassified. Similarly to that, our autoencoder will learn to put the characters to the cinematically common placements, but gonna be really weirded out once it encounters something hosed up or purposefully weird. Other than that, though, there's the whole "comprehension" versus "chinese room" problem; a lot of the neural net poo poo can look super impressive but if you go back and analyse the results and toy with some different or purposefully weird testing data, you quickly realize that it's less that you have some sorta amazing AI on your hands, and more that the problem you're presenting it with was actually loving stupid, and that it found out a really cheap way to pretend it comprehends while it's just a smooth brute force.
|
|
#
?
Dec 28, 2017 00:58
|
|
|
Teal posted:It would be easier to formulate the differences in exact and specific terms if we had complete understanding of how our mind and memory works. this is not true max-pooling gives invariance to a bunch of things assuming you've trained correctly that said humans don't have rotation invariant facial feature detectors: look at the thatcher illusion you don't need to be perfect to replace a bunch of repetitive jobs esp if you can control for things like rotation and scaling using such tricks as "lenses" and "orienting" like all the fruit classification systems do
|
|
#
?
Dec 28, 2017 01:35
|
|
|
The future is now
|
|
#
?
Dec 28, 2017 04:23
|
|
|
Malcolm XML posted:this is not true That's a bit of a simplification; it does give decent spatial invariance in position, but if the position actually matters to you, or if some complex relation of the positions matters to you (for example, text), you'll have to figure out various hacky stuff that makes the whole deal often harder and harder to train. Yeah, the U model is all the rage, but I don't think it's quite where we gonna end at. quote:you don't need to be perfect to replace a bunch of repetitive jobs esp if you can control for things like rotation and scaling using such tricks as "lenses" and "orienting" like all the fruit classification systems do Yeah that's true, again, I'm less critiquing usefulness and practical use of today NNs and more of trying to explain (in probably quite badly worded way) how are not quite there when it comes to the idea of them dreaming up movies at the drop of the hat (not even conceptually yet).
|
|
#
?
Dec 28, 2017 11:43
|
|
|
Owlofcreamcheese posted:The future is now Is this the leaked villain from the next Bond movie? Cybernetic prosthesis and a pet duck?
|
|
#
?
Dec 28, 2017 11:46
|
|
|
AI Colors/Recolors Images And of course the examples are anime girls.  
|
|
#
?
Dec 29, 2017 20:18
|
|
|
That's not exactly an impressive example considering there's no basis in any realism and any vivid color at all will work. It just needs a not-terrible color palette and it can put pretty much any color anywhere for an acceptable final result. Please don't post anime.
|
|
#
?
Dec 29, 2017 21:33
|
|
|
The left side version of the second image is better, the blue right side image have lost some deep feel and everything is kind of ...on foreground all the time. A algorithm did this? woot!
|
|
#
?
Dec 29, 2017 21:43
|
|
|
Wow, that's impressive. Seems like something like this would make it waaayyy easier to do colored manga.
|
|
#
?
Dec 29, 2017 22:08
|
|
|
Cicero posted:Wow, that's impressive. Seems like something like this would make it waaayyy easier to do colored manga. And that sounds like a much harder problem
|
|
#
?
Dec 30, 2017 02:04
|
|
|
Anime is good.
|
|
#
?
Dec 30, 2017 03:32
|
|
|
Is there some kind of anime-related corollary to the rule about new technology and porn?
|
|
#
?
Dec 30, 2017 03:42
|
|
|
Stairmaster posted:Anime is good. That sounds like something an AI would say
|
|
#
?
Dec 30, 2017 03:52
|
|
|
Taffer posted:Please don't post anime. I mean, if the algorithm is behind it, I'm interested. Also, the cold fact is that anime and manga is likely where we'll see the effects of GAN-based automation first or at least very early on (the anime industry is notoriously miserly, even though it costs less to make a Japanese toon than an American or French one), hence why I mentioned it so much in my original post, so I dunno what to say. Those cheap-rear end anime studios are probably diverting what few cents they have to spare towards funding this sorta stuff anyway since they physically can't force animators to work 25 hours a day.
|
|
#
?
Dec 30, 2017 04:30
|
|
|
withak posted:Is there some kind of anime-related corollary to the rule about new technology and porn? Already too much overlap as it is.
|
|
#
?
Dec 30, 2017 06:38
|
|
|
Cicero posted:Wow, that's impressive. Seems like something like this would make it waaayyy easier to do colored manga. (Most) manga still uses traditional media. Tones and inking are done by hand for style, not for efficiency or cleanliness. Some artists choose to work digital, or work partly in digital, but the majority are traditional. Younger artists are moving now towards digital, but it does lose some of the "feel" with how much cleanliness digital can have. On the subject of generative design: I'm an industrial designer, and while I I am stoked to see where it takes us, I caution notions of "this will replace designers." Design is not just the aesthetic approach. Designers still need to begin with indepth research and observation, and need to understand the issues to solve and the constraints -- this is a region that really, really needs more efficiency (way more so than the iterative portion). Designers are also necessary for setting up those constraints, and curating/tweaking results so that the most appropriate thing is chosen (eg making sure it fits within a brand's design language, cultural considerations). However, generative design looks like it'll be excellent at quickening the iterative portions, especially for detail variations. A designer will still need to carry out user testing and again, retweaking constraints. And a designer will still need to have a background in understanding human factors, ethnography, usability, and aesthetics. This is a good example: https://gallery.autodesk.com/fusion360/projects/elbo-chair--generated-in-project-dreamcatcher-made-with-fusion-360 (though personally, I find that chair pretty ugly). I can talk more to design disciplines if anyone would like (though again, my focus is in industrial/product, so I can't speak as much to interaction, graphic, or architecture).
|
|
#
?
Dec 30, 2017 09:22
|
|
|
If you get nightmares from watching Boston Dynamics videos: the new Black Mirror season has an episode about SpotMini relentlessly hunting a woman through a post-apocalyptic landscape with a kitchen knife. It's the one called "Metalhead". Enjoy!
|
|
#
?
Dec 30, 2017 14:02
|
|
|
*clears throat* "Self - driving cars."
|
|
#
?
Dec 30, 2017 15:30
|
|
|
Stairmaster posted:Anime is blood. Yes, anime *is* blood.
|
|
#
?
Dec 31, 2017 22:50
|
|
|
Awful news for all of us who hoped their tomato picking job would be safe from automation. The presented success rate of 75% isn't that impressive though. How often do you guys drop a tomato you're attempting to pick, guys? Teal fucked around with this message at 17:33 on Jan 1, 2018 |
|
#
?
Jan 1, 2018 17:30
|
|
|
Teal posted:Awful news for all of us who hoped their tomato picking job would be safe from automation. 75% could be really great depending on the speed
|
|
#
?
Jan 1, 2018 17:45
|
|
|
Just pay one person to scurry around the base of the robot catching falling tomatoes with a net. Then make a retro video game about it.
|
|
#
?
Jan 1, 2018 17:53
|
|
|
Teal posted:Awful news for all of us who hoped their tomato picking job would be safe from automation. IMO that's the wrong way to look at it. If now you only need to hire people who walk around and pick up tomatoes that the machines dropped, you can cut your workforce by 90%, without affecting productivity. Out of 100 workers, that's 90 workers who have been displaced by automation and need to find new jobs demanding equal or lower qualifications and skills, but offering same or higher pay
|
|
#
?
Jan 1, 2018 17:54
|
|
|
The camera and the software also look like the thing they are demoing, the actual claw that grabs looks like it was 3D printed on a printer from 9 years ago with the little low resolution layer ridges visible. I imagine they are waiting for some giant farming equipment company to call them up and say "we saw your demo, we'd like to integrate this into our type D so and sos, is this compatible?" and then at that point they will make the actual physical mechanism a real thing. This looks like real software and then 'you get the idea for this demo" hardware. Like you see them drop one in the video and you can instantly think of the fix and so can they, but the mechanism is just a proof of concept type thing, you see how it'd work. Owlofcreamcheese fucked around with this message at 18:01 on Jan 1, 2018 |
|
#
?
Jan 1, 2018 17:57
|
|
|
Doc Hawkins posted:Just pay one person to scurry around the base of the robot catching falling tomatoes with a net. Then make a retro video game about it. Then make it an online game, connect it to a tomato-catching robot, and fire the one person.
|
|
#
?
Jan 1, 2018 18:03
|
|
|
Guys you're reading too much into the thing I said about dropping, I just really wanted to make a post where I get to say both tomato and automation
|
|
#
?
Jan 1, 2018 19:26
|
|
|
Teal posted:Guys you're reading too much into the thing I said about dropping, I just really wanted to make a post where I get to say both tomato and automation You missed your opportunity to say tomatomation. Or automatoes. I'm morbidly curious to see what's going to happen if automation starts picking away at low wage jobs primarily done by migrant labor. That's the kind of thing that most Americans aren't ever going to notice because they already don't understand (or care) how their tomatoes or whatever get to them so cheaply. gently caress with the jobs of noble truck drivers and huge swathes of the country will go into a panic, but it's hard to imagine anyone actually caring about crop pickers.
|
|
#
?
Jan 1, 2018 19:46
|
|
|
Paradoxish posted:You missed your opportunity to say tomatomation. Or automatoes. My state governor appears to care about crop pickers because they're so important to the state economy.
|
|
#
?
Jan 1, 2018 19:59
|
|
|
LLSix posted:My state governor appears to care about crop pickers because they're so important to the state economy.
|
|
#
?
Jan 1, 2018 20:07
|
|
|
nier automatoes
|
|
#
?
Jan 1, 2018 20:12
|
|
|
A Buttery Pastry posted:Important for the businesses he cares about, or important because of the money they spend? Gonna lead to pretty different responses to automatoes. Very few state governments care much about money spent by crop pickers. The pay is often less than minimum wage, and frequently a large percentage of even that small amount gets sent to the worker's family in Mexico anyway. As such if a governor cares about it she cares about the businesses being affected generally.
|
|
#
?
Jan 1, 2018 21:07
|
|
|
I don't think it's going to totally change everything or anything like that but I bet all the various farm assist technology are gonna lead to some really cool stuff. Like big giant factory farms are going to still dominate everything but the more a single person and a bunch of low cost helper robots can do the more democratized small scale farming can be. Like I don't know anyone that would go off the grid and truly change their lifestyle all around being a yard farmer or anything but I absolutely know a ton of people that already garden on a small scale and would absolutely be down with scaling up if someone sold some sort of farming roomba that automated some portion of the day to day tasks. Like a drone that flew around every day and chopped up a few weeds the best it could or picked bugs off leaves or kept stuff watered or whatever. The boring parts of gardening. Like I don't think many people would convert their whole property to self sufficient corn fields or anything, but I bet a lot of people would be totally way into giving part of their lawn to fresh fruits/vegitables if cheap enough robots got it to the point it was something you had to personally deal with on a once every three weeks basis instead of an every day activity. And it seems like the sort of thing that could get big as a hobby in middle class suburbs then eventually have real impact in other countries where farming is a bigger deal if the price could drop enough.
|
|
#
?
Jan 1, 2018 22:23
|
|
|
Owlofcreamcheese posted:nier automatoes For the glory of marinara
|
|
#
?
Jan 1, 2018 22:31
|
|
|
How does more and more expensive equipment help democratize farming? Historically, it has been capital intensive labor-saving machinery that has led to farm consolidation. You need a farm large enough to make owning the machine make economic sense. Renting the machine or taking a loan to buy it is the exact process that leads to small farms dying as they get squeezed between fixed or rising costs and crop prices set lower by the economies of scale massive farms present.
|
|
#
?
Jan 1, 2018 22:40
|
|
|
Owlofcreamcheese posted:I don't think it's going to totally change everything or anything like that but I bet all the various farm assist technology are gonna lead to some really cool stuff. Like big giant factory farms are going to still dominate everything but the more a single person and a bunch of low cost helper robots can do the more democratized small scale farming can be. Like I don't know anyone that would go off the grid and truly change their lifestyle all around being a yard farmer or anything but I absolutely know a ton of people that already garden on a small scale and would absolutely be down with scaling up if someone sold some sort of farming roomba that automated some portion of the day to day tasks. Like a drone that flew around every day and chopped up a few weeds the best it could or picked bugs off leaves or kept stuff watered or whatever. The boring parts of gardening. I'm in favor of automating farming further, but you are woefully unaware of the state of that industry if you think it democratizes it. These are not "farming roombas", they're gigantic industrial machines made to work in enormous volumes. To start a farming operation even for a single crop you're looking at several million dollars in equipment if you want state-of-the-art automation, which this is. And that's not even counting land requirements. It very effectively pushes out small players and ensures that only the big players with mountains of capital can stay competitive.
|
|
#
?
Jan 1, 2018 23:11
|
|
|
Taffer posted:I'm in favor of automating farming further, but you are woefully unaware of the state of that industry if you think it democratizes it. These are not "farming roombas", they're gigantic industrial machines made to work in enormous volumes. To start a farming operation even for a single crop you're looking at several million dollars in equipment if you want state-of-the-art automation, which this is. And that's not even counting land requirements. I'm taking millennials with victory gardens. Not the return of family farms. People growing a small percentage of their food because it's made easy to do so. Not people dropping out to run farms.
|
|
#
?
Jan 1, 2018 23:18
|
|
|
Owlofcreamcheese posted:The boring parts of gardening. The people that could afford to spend money on a Garoomba already do this with gardeners.
|
|
#
?
Jan 1, 2018 23:21
|
|
|
Owlofcreamcheese posted:I don't think it's going to totally change everything or anything like that but I bet all the various farm assist technology are gonna lead to some really cool stuff. Like big giant factory farms are going to still dominate everything but the more a single person and a bunch of low cost helper robots can do the more democratized small scale farming can be. Like I don't know anyone that would go off the grid and truly change their lifestyle all around being a yard farmer or anything but I absolutely know a ton of people that already garden on a small scale and would absolutely be down with scaling up if someone sold some sort of farming roomba that automated some portion of the day to day tasks. Like a drone that flew around every day and chopped up a few weeds the best it could or picked bugs off leaves or kept stuff watered or whatever. The boring parts of gardening. yea I want this. a ~1TEU sized greenhouse with something like a UR10 mounted to a rail down the middle. it doesn't even have to be that well scripted, if I could just remotely pilot it like an even more boring version of farmville while at work.
|
|
#
?
Jan 1, 2018 23:23
|
|
|
|
| # ? May 8, 2024 13:29 |
|
|
NerdyMcNerdNerd posted:The people that could afford to spend money on a Garoomba already do this with gardeners. I am sure someone has done that in the history of the world or whatever but I have absolutely never heard of someone hiring gardeners so they can grow their own garlic in a home garden. I'm talking about something like this but not so obviously made of 3D printers and arduinos. https://www.youtube.com/watch?v=zXplJ3Cb2pQ Probably either built into the wooden frame or just a drone not connected to anything so it's just a thing you can buy at target then not need to buy onions again. Something that isn't going to even make an attempt to generate 2000 calories a day and make a person a farmer but something that might create the number of raspberries your family eats per year.
|
|
#
?
Jan 1, 2018 23:37
|
|