Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[alway]

I was reading an article a while ago about image-sharpening via AI. Some commenters then were saying it's impossible, since you can't create information that wasn't in the source image, therefore there can't possibly be clearer detail in the outcome image. What those commenters miss is that the AI can bring in outside knowledge of how objects work, and thus synthesize the missing details. e.g. if you have a fuzzy face photo, then the algorithm can compute known details of the face, combined with what it knows about how faces work in general, and "sharpen" the image beyond the amount of information that's present in the fuzzy photo. You could actually achieve this quite simply, by taking millions of mugshots, fuzzing them up, then training a neural network to extrapolate the originals from the fuzzy versions.

I think that the stuff in that article is along similar lines, using very large training sets. It's potentially very interesting if you could just turn text into images. For instance, imagine a system that automatically illustrates existing stories based on text. All sort of books, text adventures and the like could theoretically have illustrations automatically generated.

As for the sharpening of images, care must be taken as to what it's used for, since it's a case of "Yes, but..."
Yes, it does result in a sharper images, clued in by the original image... But, the details are essentially imagined based on what the training data suggests could be plausible. For purposes of identification, like cleaning up grainy surveillance footage to ID a suspect, it becomes very dangerously misleading, since the added detail is nothing more than the most plausible subset of all possible details. If you've ever seen someone from a distance and thought you recognized them, then got closer and realized the face details were off despite a similar overall structure, and it was actually just a stranger? That sort of mis-identification applies to this as well.

[Shadowlord]

That's what I'd be concerned about too. Facial identification software is already prone to making mistakes.

[Reelya]

Well the best you can hope for is that they make mistakes in the same sort of range of situations that a human does and then hopefully bring in more cues to minimize the errors.

[Starver]

As I buried it within a large amount of other chatter, here's my main point again.

We can't he [be, sic] sure that we could make it better than a siitably [suitably] trained/experienced human, just more consistent regardless of caffeine (and/or ethanol) levels, etc... 

(I only noticed my errors after I got replies. To err is human, of course. But to get things consistently wrong might take a computer. )

[Reelya]

https://hardware.slashdot.org/story/17/01/19/2332242/neuroscience-cant-explain-how-a-microprocessor-works

The Economist has an interesting story about two neuroscientists/engineers -- Eric Jonas of the University of California, Berkeley, and Konrad Kording of Northwestern University, in Chicago -- who decided to test the methods of neuroscience using a 6502 processor. Their results are published in the PLOS Computational Biology journal. Neuroscientists explore how the brain works by looking at damaged brains and monitoring inputs and outputs to try to infer intermediate processing. They did the same with the 6502 processor which was used in early Atari, Apple and Commodore computers. What they discovered was that these methods were sorely lacking in that they often pointed in the wrong direction and missed important processing steps.

It'd be interesting if such work can result in new ideas for neuroscience itself.

[wierd]

A brief examination of the "Chinese room" thought experiment could have told you that the kind of research that was being done prior to this experiment was very faulty.

They would be much better served by trying to make very very large neurons that they could hook all kinds of probes to.

[Sheb]

A brief examination of the "Chinese room" thought experiment could have told you that the kind of research that was being done prior to this experiment was very faulty.

They would be much better served by trying to make very very large neurons that they could hook all kinds of probes to.

Both approaches help. The brain is such a complex thing that you can only go so far by starting with single neurons.

(BTW, we do have huge neurons that we use to stick probe into. The Squid Giant Axon can measure up to 1 mm of diameter, and has been used for early research on neurons.

[Reelya]

I think it's deeper than that, the tools that a neuroscientist has are meant to be able to work out how an unknown structure that does some processing is doing it's job. Sure we have a lot of rules of thumb specifically related to human behavior, but the fundamental tools of neuroscience should be able to be used to devise ways to deduce how any brain-like structure operates.

We aren't starting from a position of understanding how the brain works, they're meant to work that out with their tools. If their tools fail to work out how a very simple chip is doing it's processing then that looks bad for the tools as correctly working out how a much more complex unknown structure is operating.

Also, consider that the tools that a neuroscientist has should ideally be tools that you could use to determine how e.g. an exobiological brain operates, or brains from very different creatures to ourselves. If they can't even deduce how a 6502 processor works then there's little hope in them using the neuroscience tools to e.g. work out how an ant or bee brain works, let alone answering deep questions such as how consciousness forms.

[Parsely]

-snip-

Thank you for sharing that, really fun read!

[Reelya]

This material sounds kinda fun:
https://science.slashdot.org/story/17/01/10/2355259/mit-unveils-new-material-thats-strongest-and-lightest-on-earth

Graphene, which was heretofore, the strongest material known to man, is made from an extremely thin sheet of carbon atoms arranged in two dimensions. But there's one drawback: while notable for its thinness and unique electrical properties, it's very difficult to create useful, three-dimensional materials out of graphene. Now, a team of MIT researchers discovered that taking small flakes of graphene and fusing them following a mesh-like structure not only retains the material's strength, but the graphene also remains porous. Based on experiments conducted on 3D printed models, researchers have determined that this new material, with its distinct geometry, is actually stronger than graphene -- making it 10 times stronger than steel, with only five percent of its density. The discovery of a material that is extremely strong but exceptionally lightweight will have numerous applications. As MIT reports: "The new findings show that the crucial aspect of the new 3-D forms has more to do with their unusual geometrical configuration than with the material itself, which suggests that similar strong, lightweight materials could be made from a variety of materials by creating similar geometric features."

Basically graphene is individual carbon atoms aranged in a honeycomb pattern, so now they're putting layers of that stuff together to make even stronger materials. It's basically atom-scale carbon chainmail.

[Loud Whispers]

GRAPHENE FASHION WEN

[Loud Whispers]

I suppose now is not a good time to mention my new line of asbestos scarves

[Reelya]

There are already a number of alternatives to graphene being looked at e.g. Silicene and Phosphorene. This area is really new. Silicene was first identified in 2010, and Phosphorene in 2014. Basically there's a whole unexplored world of atom-scale material engineering opening up. And that's before we even get into the possibilities of mixing up materials in specific patterns.

[martinuzz]

Whaddaya mean 'you shouldn't grow your weed on asbestos wool'?

[inteuniso]

GRAPHENE FASHION WEN

Good distance down the line. Apparently graphene is a major carcinogen, since it can flake, and when it does so you wind up inhaling little bits, and those little bits are so thin they can just straight up enter your cells and mess with your DNA.

The major problem with them right now is that graphene is made using chemical vapor deposition onto a metal substrate, or a plasma furnace using metal substrates. This leads to major impurities & metals being attached to carbon, which otherwise shouldn't be a problem.

It will also oxidize if it's not functionalized properly, and when the carbon is used for biological purposes the oxygen is freed and the oxygen becomes a carcinogenic free radical.

Fortunately it can be functionalized with chemical exfoliation using an organic acid, 2(COOH). This can be plant-derived so it's ultra-inexpensive as well. This is usually done in an aqueous solution and once functionalized, graphite clumps together, and it stops being so flaky.

So, to answer the when for fashion... later this year. I'm opening up the indiegogo tomorrow for a full patent and I'll be offering graphene-dyed (actually more of a mordant, let me know what color you want and I'll start exploring the various plants that allow for cheap production of that color dye)  hemp fabric clothing, to be delivered by september. Just waiting to put up the campaign because I want to test my dyeing skills out (if I dye properly, it should be fairly bulletproof, I'll record a video).