Group items tagged

from one of the reddit threads discussing this: "bit fishy, crazy if real". "Incredible claims: - Train only using about 10% of imagenet-12, i.e. around 120k images (i.e. they use 6k images per arm) - get to the same or better accuracy as the equivalent VGG net - Training is not via backprop but more simpler PCA + Sparsity regime (see section 4.1), shouldn't take more than 10 hours just on CPU probably "

clicking the link says the manuscript was withdrawn :))

This "one-shot learning" paper by Googe Deepmind also claims to be able to learn from very few training data. Thought it might be interesting for you guys: https://arxiv.org/pdf/1605.06065v1.pdf

Interesting mater thesis that lead to the just announced new launcher ! Remember our discussion after the advanced concepts workshop!?!?!

do you know about this? https://www.qb50.eu/launch.php

I knew about the project only, but not about this very interesting missile ... "The launcher has a remarkable track record of several hundred successful launches and only 1 failure. Shtil is marketed by the State Rocket Center Makeyev."

The end is near...

che brutta figura ...

Some quotes: "Our results are in agreement with what Einstein would like to have" - they called him just to double-check... "How many times do you have to say 'zero' to make sure it's zero?" - because statistics is just not trendy any more...

and what is your take?

As William Gibson famously said "The future is already here - it's just not very evenly distributed." Many of the listed technologies are already here, in one way or another, the authors are just putting dates on when they think they'll achieve greater consumer impact. I have issues with many of those dates. I think there are a lot of under- and overestimations. Also, what a bleak future for AI it that's all it'll give the world in the next 30 years!

Find yourself in an area without breathable air? No problem... just use this!

Quote: Recognizing the need to investigate and evaluate visionary aerospace concepts for their validity, several organizations employ specialized groups to do just this, like ESA's Advanced Concepts Team [2], Lockheed Martin's Skunk Works [3] or NASA's Team-X [4].

hmmm, I was wondering how many ESA employees do know that ACT does exist....

And my son studying at Stanford (he just sent me the same link !) follows the courses this semester of two of the teachers mentioned in the article, Thrun - very good and Ng - excellent

A new class of 2D semiconductor has been developed by researchers in the US. The free-standing quantum membranes, which are made of indium arsenide (InAs) and are just a few layers thick, have properties that are in stark contrast to the semiconductor layers used in conventional transistors.

Just in time for the end of the winter ....

Mantis shrimp seem to have 12 types of photo-receptive sensors - but this does not really improve their ability to discriminate between colors. Speculation is that they serve as a form of pre-processing for visual information: the brain does not need to decode full color information from just a few channels which would would allow for a smaller brain. I guess technologically the two extremes of light detection would be RGB cameras which are like our eyes and offer good spatial resolution, and spectrometers which have a large amount of color channels but at the cost of spatial resolution. It seems the mantis shrimp uses something that is somewhere between RGB cameras and spectrometers. Could there be a use for this in space?

> RGB cameras which are like our eyes ...apart from the fact that the spectral response of the eyes is completely different from "RGB" cameras (http://en.wikipedia.org/wiki/File:Cones_SMJ2_E.svg) ... and that the eyes have 4 types of light-sensitive cells, not three (http://en.wikipedia.org/wiki/File:Cone-response.svg) ... and that, unlike cameras, human eye is precise only in a very narrow centre region (http://en.wikipedia.org/wiki/Fovea) ...hmm, apart from relying on tri-stimulus colour perception it seems human eyes are in fact completely different from "RGB cameras" :-) OK sorry for picking on this - that's just the colour science geek in me :-) Now seriously, on one hand the article abstract sounds very interesting, but on the other the statement "Why use 12 color channels when three or four are sufficient for fine color discrimination?" reveals so much ignorance to the very basics of colour science that I'm completely puzzled - in the end, it's a Science article so it should be reasonably scientifically sound, right? Pity I can't access full text... the interesting thing is that more channels mean more information and therefore should require *more* power to process - which is exactly opposite to their theory (as far as I can tell it from the abstract...). So the key is to understand *what* information about light these mantises are collecting and why - definitely it's not "colour" in the sense of human perceptual experience. But in any case - yes, spectrometry has its uses in space :-)

Speed is blinding. Just ask the tiger beetle: This predatory insect has excellent sight, but when it chases prey, it runs so fast it can no longer see where it's going.

By just getting a DNA footprint of a person scientists (and soon police) can produce an image of the person's face. Check out the pictures!

should try it the other way around, deduce the DNA from facial features. That would be even cooler.

Well actually, they did something like that as they searched for common DNA patterns in people that had similar facial features. With a large enough dataset that could provide already 24 DNA tracers that could used reliably for prediction. Imagine if you had even more data available, who needs a model then... just let the NN do it :)

theoretical physicists... :) Read the last sentence of the paper...in this way anyone can publish in nature...just make a good story with little evidence Did dark matter kill the dinosaurs? The Solar System's periodic passage through a 'dark disk' on the galactic plane could trigger comet bombardments that would cause mass extinctions.

Hmm.. right.. then again, this is not an actual journal publication but a news broadcast. But you are right that the name Nature is attached to it so the journal is definitely banking on their acquired status.

why are we not getting these type of startups in Europe .... btw: Will is british

I agree, the mentality is different, it's hard to find VC funding for hardware stuff, even more so if you wanna shoot your HW into space. but there is movement, e.g. pioneers.io, they are in vienna and are actively trying to get more VC funding (in europe) for HW and other engineering startups

thanks for the link ... just read this blog ... http://pioneers.io/blog/space-race-2-0-putting-satellites-into-the-hands-of-everyone a lot of selling talk but fundamentally I agree that they have a point ... and as ACT we will face the criticism in not so long that we have not managed (nor tried hard enough) to convince ESA about the need to embrace this "new space"

Oh those bloody physicists... someone forgot to turn off the lights and now they have a 400 times better conductive crystal. If science was always that easy, I would light a candle every day.

Apparently this was not the first discovery of this effect involving SrTiO3. In an article from 2012 a conductance increase of 5 orders in magnitude is described (http://pubs.acs.org/doi/abs/10.1021/nn203991q). But indeed many large impact discoveries are accidental...

I thought we all knew already that science is just another form of directed random search :)))

A fake paper was accepted in 60% of tried open access journals, just because the author paid the fee :)

Love the comment that some papers which originally rejected it, in the end did accept it anyway after some correspondence

Does this have implications for AI algorithms??

I was actually thinking exactly about this question during my bike ride this morning. I am surprised that some codons would need to have a double meaning though because there is already a surplus of codons to translate into just 20-22 proteins (depending on organism). So there should be about 44 codons left to prevent translation errors and in addition regulate gene expression. If - as the article suggests - a single codon can take a dual role, does it so in different situations (needing some other regulator do discern those)? Or does it just perform two functions that always need to happen simultaneously? I tried to learn more from the underlying paper: https://www.sciencemag.org/content/342/6164/1367.full.pdf All I got from that was a headache. :-\

Probably both. Likely a consequence of energy preservation during translation. If you can do the same thing with less genes you save up on the effort required to reproduce. Also I suspect it has something to do with modularity. It makes sense that the gene regulating for "foot" cells also trigger the genes that generate "toe" cells for example. No point in having an extra if statement.

beati pauperes spiritu :-) "Political junkies might be interested to learn that conservative women are particularly blissful: about 40 percent say they are very happy. That makes them slightly happier than conservative men and significantly happier than liberal women. The unhappiest of all are liberal men; only about a fifth consider themselves very happy." I think that there are many factors missing there that are just not asked possibly such as altruism as a source for happiness ...

According to The Beatles "Happiness is a warm gun"... To each his own, I guess. :-)

It really depends on one's values really. If your values don't follow the standard vanilla flavor then it all breaks down.

Giant sun mirrors were installed to light up one Norwegian town in winter, just to improve the mood of its residents!