Science/Technology link roundup, June 2012
Japanese biologists grow a human eye precursor from stem cells. Not yet a functioning eyeball, but an embryonic proto-eyeball structure called an "optic cup". We've long known that in principle the structure of all the body's systems and organs is encoded in the DNA of a single cell, but this is the first time that a complex three-dimensional structure has been grown on its own from human cells. A future where we can grow replacement organs from our own DNA is going to be pretty cool.
Scientists in Long Island have mapped out the "wiring" of the mouse brain. This is not the same as knowing what every neuron in the mouse brain does. It's the equivalent of sequencing the genome, which doesn't tell us what every gene does but does provide the vital high-level framework and context for future exploration. This is the first vertebrate brain to be mapped at this level of detail and is considered a first step towards mapping the human brain. The team has made lots of hi-res images publicly available.
Google Research has trained a 9-layer deep neural network to recognize faces based on an unlabeled data set. The cool thing here is that the training data is unlabeled. Usually when you train a classifier to recognize faces, you have to give it a set of pictures with faces and a set of pictures without faces, so it can learn what to look for. In this research, Google just fed the neural network thousands of pictures without any labels and it learned to tell features apart, without any knowledge of what it was supposed to be looking for. It didn't just learn to recognize faces; it also learned to recognize several other common picture elements, such as the presence of cats.
Meanwhile, Microsoft has been using the same sort of "deep neural network" method to train a system for better speech recognition; they're using it to make audio files searchable, which is pretty cool. I seem to recall my machine learning course at U of C teaching us that a neural network with lots of hidden layers doesn't perform any better than a neural network with just an input and output layer. Apparently the "deep training" Google and Microsoft are doing is based on a breakthrough that was made in 2006 -- the year after I graduated. Ha!
Google is working on augmented-reality glasses, too. That might be neat!
They are also releasing some sort of mystical black orb of doom that costs $300. But what does it do? Something music-related? I'll let the designer explain:
“The sphere is a zero primitive form,” Jones says. “It transcends into this third wave of electronics where the interface, Android, is on another device. So now the actual object doesn’t have the burden of direct manipulation. It can have any presence and gesture within the room, and this encourages you to interact with it.”
Thanks, that sure clears things up. I'm sure everybody will want one.
I had been vaguely aware that there existed proprietary pre-internet networks in Europe, but I had never heard about "Minitel" until I read this article about the rise and fall of France's own government-sponsored proto-internet service, which is finally going offline this year.
Do-it-yourself surveillance drones, because why should the military have all the fun?
Germany is trying to go completely green energy, shutting down its nuclear plants while aiming to cut greenhouse gases 40% by 2020. They describe this extremely ambitious plan as "Energiewende", an energy revolution.
A new paper by a group of 22 climate scientists and ecologists summarizes what we know on tipping points in ecosystems. You may have seen this research reported on various news sites with sensationalistic and attention-grabbing headlines like "WE'RE ABOUT TO PUSH THE EARTH OVER THE BRINK".
The actual paper, as far as I can tell, does not appear to make this claim; from what I've read it seems to be more like "our ecosystem models are prone to rapid phase transitions when they cross a tipping point; the earth might be too, so watch out". They do point out that human activity uses 43% of the earth's land surface and speculate that maybe something scary will happen when we pass 50%? They don't claim to know the answer. We might be approaching a tipping point, but we don't know when, or what it will be, or what the ecosystem will look like afterwards. It's all about uncertainty and the need to know more about how the global ecosystem works.
Unfortunately, no matter how nuanced scientists try to be in their actual statements, the media always turns it into "WE'RE ALL GONNA DIE". This is a good reason to be skeptical of science reporting in mainstream media and to try to get as close as you can to the original research (unfortunately, the actual article in question is behind a paywall.)