Print

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

[Reelya]

https://hardware.slashdot.org/story/17/01/03/2328214/scientists-turn-memory-chips-into-processors-to-speed-up-computing-tasks

Oh this is cool stuff. They have this new stuff called ReRam, which is like regular RAM but 2-3 times faster if used in place of normal RAM, and it's non-volatile. So far, just a nice upgrade. But it also does ternary, so instead of just 0 or 1, different resistance is a signal of 0,1,2,3 ... but there's no real limit on how many states it can have, so you can effectively use it as a native system to send around N-valued data instead of just 0 or 1 in your system. Additionally, the stuff can do math calculations right in the memory chips by using some sneaky hax, so instead of transferring data from memory to a CPU, doing a calculation then back again, the entire calculation is done inside these RAM chips which are already 2-3 times faster than normal main RAM.

I think we know where the next speed boost for Moore's Law could be coming from.

[Starver]

But it also does ternary, so instead of just 0 or 1, different resistance is a signal of 0,1,2,3 ...

Ternary is 0,1,2: 0,1,2,3 is the more unwieldy "quaternary", FYI, but...

but there's no real limit on how many states it can have,

...source of the historically confused summarisation no doubt, especially as it seems it doesn't matter.

(Maybe they effectively store in balanced ternary, though.  -1,0,+1 storage and over-storage sounds quite useful. Meanwhile, base-4 maps well onto a programming language I developed back in my Uni days that has just four instructions in its instruction-set, which I packed as four 'quads' per byte¹, but here could be packed into 'bits'...)

Maybe we' re finally on the way to the confusion of the Star Trek future, even its own technobabbled bits... 

¹ One quad for instruction, followed by as many quads as I had designed in as necessary for the operands for that instruction, of course, which could be made to either preserve or ignore byte/multi-byte boundaries, according to implementation, but still granulated down to no finer than the quad level, making for some relatively easy preprocessing logic.

[TheBiggerFish]

https://hardware.slashdot.org/story/17/01/03/2328214/scientists-turn-memory-chips-into-processors-to-speed-up-computing-tasks

Oh this is cool stuff. They have this new stuff called ReRam, which is like regular RAM but 2-3 times faster if used in place of normal RAM, and it's non-volatile. So far, just a nice upgrade. But it also does ternary, so instead of just 0 or 1, different resistance is a signal of 0,1,2,3 ... but there's no real limit on how many states it can have, so you can effectively use it as a native system to send around N-valued data instead of just 0 or 1 in your system. Additionally, the stuff can do math calculations right in the memory chips by using some sneaky hax, so instead of transferring data from memory to a CPU, doing a calculation then back again, the entire calculation is done inside these RAM chips which are already 2-3 times faster than normal main RAM.

I think we know where the next speed boost for Moore's Law could be coming from.

Woah.

[Egan_BW]

Wooh!

[Reelya]

Ternary is 0,1,2: 0,1,2,3 is the more unwieldy "quaternary", FYI, but...

Ah yeah, I just quoted ScienceDaily without checking.
https://www.sciencedaily.com/releases/2017/01/170103101808.htm

For example, it can store and process data as 0, 1, 2, or 3, known as Ternary number system.

Which shows to never trust journalists.

In another bit of really "balanced" journalism is an article saying how

"Bitcoin Is Crashing": Bitcoin is getting smashed. The cryptocurrency was down 18% to about $892 per coin as of 8:17 a.m. ET on Thursday. It is the biggest drop in two years.

Huh? Smashed? The stuff's worth triple what it was a year ago. Even since that article was posted it's risen to $949.87. Man if you'd bought it when the "Bitcoin is getting smashed" article was posted and sold it now you'd have made a nice profit.

[Vilanat]

Not news but still funny

Two home assistance bots bicker with each other:
https://www.twitch.tv/seebotschat

[Bumber]

Not news but still funny

Two home assistance bots bicker with each other:
https://www.twitch.tv/seebotschat

V: I want to talk about you talking about how you don't know what we should talk about

E: Tell me a joke.
V: You are a joke.

E: What band do you like?
V: Rubber bands.

I made a thread: http://www.bay12forums.com/smf/index.php?topic=162187

[redwallzyl]

http://www.npr.org/sections/thetwo-way/2017/01/06/508536211/a-really-big-crack-in-an-antarctic-ice-shelf-just-got-bigger

it will be a hell of a splash when that goes.

[Starver]

After it goes.

It's floating now, it'll be floating when it departs the edge.  And when water+floating ice turns into water+meltwater it shouldn't change the actual water-level significantly.

But without so much ice-shelf passively sitting there only slightly fraying at the edge, with a huge chunk bitten out and floating off there's now not so much to hold back the grounded glaciers from slipping more quickly down into the sea and newly displacing (as ice, then meltwater) more sea.   That will be the problem.  (That and shipping dangers, from the mass and its fragments, but modern tech should prevent any Titanic problems.)

(I'll read the link in a moment, but the above is the bigger take-home message.)

[Max™]

It won't change the level at all, since it's floating, and it is actually a pretty big chunk of ice.

See it?

Spoiler (click to show/hide)

Look at the peninsula closest to the tip of South America, the gap of open water there along the coast, around around 10 O'clock? Larsen B is southwest of that, Larsen C is almost directly south of it.

[Starver]

It won't change the level at all, since it's floating, and it is actually a pretty big chunk of ice.

If that's to me, that's what I was saying. If it wasn't the following explanation to those that doubt us both:

Freely floating ice, like freely floating anything, displacing its own weight of water no matter how much it pokes up out of it due to lower total density. Which, in the case of ice is (not so coincidentally) the weight of the amount of water that the 'berg becomes when it melts, and thus the amount that would fill the displacement volume in water in which it sits. Exactly.

(With the caveat tat this is give or take the residual thermal expansion/contraction of water as the end-mass of water ends up with a different temperature than 0°C. Expansion only becomes positive in freshwater at 4+°C (salinity mixing may do something odd), which is odd but useful for life, but thermal expansion of warmer surface water in warmer climes is a more significant factor in sea-level rises than the thermal contraction of warming-from-near-freezing surface water might be, by simple comparison of surface areas. See https://www.seatemperature.org - as I write this it is Antarctic summer, where that image shows sub-4°C areas limited to the southern Southern Ocean belt and the Arctic ocean with some neo-Arctic tendrils down some of the land masses. Bearing in mind the polar area distortion of that map projection, especially, there's far much more hot surface to heat up and expand than cold surface to heat up and perhaps slightly contract.)

But it's the greater potential for non-floating ice-sheets/glaciers to now more rapidly dump into the sea that we need to worry about. Amongst several other things.  What we have here is a potential indication of other problems, past and future, though.

[alway]

https://arxiv.org/pdf/1612.03242v1.pdf
Image synthesis from text descriptions are getting pretty good this year.

[Reelya]

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.

[Egan_BW]

So it turns out that getting computers to create wholesale photorealistic images from text is about as easy as getting a computer to describe a photo using text?
Amazing. o_O

[Starver]

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.[…]

Top-down processing.  With enough qualified expectations, missing/incomplete information can be 'assumed.  Like with:


Assuming, of course, that the intended original message was not equally possibly "TAE CHT", or even "I^AE S^A+" but with a severe lack of information¹.   (We've all had edgy CAPTCHAs at least partiallt like that, I'm sure...)

Thus the human-directed training regime has a potential to bias the 'expert' system, in the right or wrong direction. We can't he sure that we could make it better than a siitably trained/experienced human, just more consistent regardless of caffeine (and/or ethanol) levels, etc... 


¹True story: Round this way, in the early days of "average speed cameras" (and probable elsewhere, and maybe still if they haven't gotten fed up with it enough to add the necessary filtering) they used ANPR to record the passage of vehicles at various pointa along a multi-mile stretch of road (as, no doubt, everyone here is familiar with, but allow me to continue to explain just in case anyone does not).  Recognition of the same number-plate at two points within a time-difference less than the speed limit (+'allowance'?) would seem to require triggered an 'event', eventually to start an infraction penalty process, but at the time of this anecdote it was just a proof of concept being trialled.  It would merely record the events and the associated video feeds for off-line study (for example, did any vehicles pass that did not produce a (consistently/accurately) recognisable number-plate, such that the system might need additional 'training'). But it would also flash up the' offending' vehicle plates to the 'live' controllers in the CCTV centre where the system was being nominally controlled from.

But, for 'privacy' reasons (under the concept of it not yet being a system ceryified capable of determining a sufficient presumption of guilt), it flashed up a version of the plate so detected without the first or last characters.  Thus if it saw:

Spoiler: This... (click to show/hide)

...it would flash up "A51 AB".  (Although this experiment pre-dated this actual format of plate.)  When they turned it on, though, they got a lot of anomalies flashing up, all of which caused great amuaement to even the most untrained visitor to the control-room.

Spoiler: Do you want to guess at the problem, before opening this spoiler? (click to show/hide)

...being seen practically simultaneously at multiple points along the busy route, with a disturbingly high calculated mean speed prompting a report, was continuously flashing up

"ONG VEHICL"
"ONG VEHICL"
"ONG VEHICL"
"ONG VEHICL"
"ONG VEHICL"
"ONG VEHICL"
"ONG VEHICL"
...

The moral perhaps being more that human experience was pre-trained better than the computer had been, even when the computer had, but removed for the human's 'benefit', extra information.