Rather than building a single model and seeing how well it could recognize visual objects, the team constructed thousands of candidate models, and screened for those that performed best on an object recognition task
the algorithm is designed for robots that will be monitoring an environment for long periods of time, tracing the same routes over and over. It assumes that the data of interest — temperature, the concentration of chemicals, the presence of organisms — fluctuate at different rates in different parts of the environment.
But it turns out to be a monstrously complex calculation. “It’s very hard to come up with a mathematical proof that you can really optimize the acquired knowledge,”
The new algorithm then determines a trajectory for the sensor that will maximize the amount of data it collects in high-priority regions, without neglecting lower-priority regions.
At the moment, the algorithm depends on either some antecedent estimate of rates of change for an environment or researchers’ prioritization of regions. But in principle, a robotic sensor should be able to deduce rates of change from its own measurements, and the MIT researchers are currently working to modify the algorithm so that it can revise its own computations in light of new evidence. “
"Switchable nanomaterials-materials that can change their properties and/or function in response to external stimuli-have potential applications in electronics,..."
Very very nice idea! Unfortunately samples show they still have quite some work to do on the image processing side... which is interesting because there are very good open-source solutions for stitching.
Whether society embraces face recognition on a larger scale will ultimately depend on how legislators, companies and consumers resolve the argument about its singularity. Is faceprinting as innocuous as photography, an activity that people may freely perform? Or is a faceprint a unique indicator, like a fingerprint or a DNA sequence, that should require a person's active consent before it can be collected, matched, shared or sold?
Actually these sort of things are also quite easy to exploit. Print a picture of Osama bin Laden on your t-shirt and have the entire police force scared out of their wits.
"The OpenBCI Board is a versatile and affordable analog-to-digital converter that can be used to sample electrical brain activity (EEG), muscle activity (EMG), heart rate (EKG), and more"
Perhaps some work or ideas on brainwave analysis would be interesting ? (User interfaces, mood classifier, detection of various alertness levels )
It's not worth it for $400... The chips are seriously nothing special and you can get a lot better for a lot cheaper. I would just get the electrodes and link them to a RPi or an Odroid or something.
True, but the selling feature here is that they take care of that stuff and sell it for 400$. Lets say the hardware is 100USD, then an RF-grade person here here has to do the coding, interfacing, testing within roughly (300/16eur/hour) 20 hours to break even and even then the interface is much nicer in their case.
Amazing how some guys from some other university also did pretty much the same thing (although they didn't use the bidirectional stuff) and published it just last month. Just goes to show you can dump pretty much anything into an RNN and train it for long enough and it'll produce magic.
http://arxiv.org/pdf/1410.1090v1.pdf
LSTMs: that was also the first thing in the paper that caught my attention! :)
I hadn't seen them in the wild in years... My oversight most likely. The paper seems to be getting ~100 citations a year. Someone's using them.
Other possible study: get a textbook example of an image of a pen, evolve it just enough so NN can't recognize it anymore, while minimizing the distance between the original and evolved images.
EDIT: Its been done already: http://cs.nyu.edu/~zaremba/docs/understanding.pdf
Of course, you can't really use them to extrapolate. The unknown unknown is always the trickiest :P
They should just make another class "random bullshit", really and dump all of this stuff in there.
I think there's a potential paper right there
Typically, we send rovers to our planetary neighbors one at a time -- but what if we sent a small team of smaller, less impressive robots instead? That's the idea NASA is exploring at Kennedy Space Center with Swarmies: a quartet of four autonomous robots designed to work together to complete a single mission.
"As in a colony of ants, "there's no centralized leader per se," Rubenstein says." Thought of ant colony document permanently on Dario's desk. Some relevance?