The Math of a Fly's Eye May Prove Useful 90
cunniff writes "Wired Magazine points us to recent research that demonstrates an algorithm derived from the actual biological implementation of fly vision (PLoS paper here). Quoting the paper: 'Here we present a model with multiple levels of non-linear dynamic adaptive components based directly on the known or suspected responses of neurons within the visual motion pathway of the fly brain. By testing the model under realistic high-dynamic range conditions we show that the addition of these elements makes the motion detection model robust across a large variety of images, velocities and accelerations.' The researchers claim that 'The implementation of this new algorithm could provide a very useful and robust velocity estimator for artificial navigation systems.' Additionally, the paper describes the algorithm as extremely simple, capable of being implemented on very small and power-efficient processors. Best of all, the entire paper is public and hosted via a service that allows authenticated users to give feedback."
Anyone know about bees? (Score:5, Interesting)
The researchers drew their algorithm from neural circuits attuned to side-to-side yaw, but O’Carroll said the same types of equations are probably used in computing other optical flows, such as those produced by moving forward and backwards through three-dimensional space.
I vaguely remember seeing a study that examined how bees travel without hitting anything but using very few neurons. Something about the relative size change of objects between eyes. They tested this by putting bees in a clear tunne with patterns on belts on the right and left walls. By changing the speed of the belts, the bees would ram into the walls, but as long as the belts were moving at the same speed, the bees were fine. Is this ringing a bell for anyone else?
Thousand scientists in a room with a typewriter... (Score:2, Interesting)
Though they built the system, the researchers don’t quite understand how it works.
and...
Intriguingly, the algorithm doesn’t work nearly as well if any one operation is omitted. The sum is greater than the whole, and O’Carroll and Brinkworth don’t know why.
Wow, some interesting "science" that's going on here.
Great result, but, really, way to go guys! You can't understand a non-linear system's behavior; join the club. I still can't understand why z_n+1 = z_n^2 + c looks so pretty either.
Extinction (Score:2, Interesting)
You know I can forsee a time, a few hundred years down the line, where we are recovering from the environmental catastrophe caused by man. In this time I think the great profiteers of the day will look down shamefully on the profiteers of today, who destroyed so many of natures feats of engineering in order to harvest lumber or food. I think that they will look back at all the diversity that could have been exploited for their designs and curse us. Mankind will probably be roaming the stars in search of biodiversity by then. If for no other reason then to get the hundreds of thousands of years benefit of tried and tested solutions to engineering problems.
Re:We don't understand it but we can do it (Score:3, Interesting)
A scientist says, "This works, but I don't know why, How do I complete the theory?"
An engineer says, "This works, but I don't know why. How do I use it to build something that does what I want?"
A good engineer says, "This works, but I don't know why. How do I use it to build something that does what I want.. And, in what domain does my model break down and how do I make sure I don't get my system into that domain?"
Sizable chunks of control theory, frequency analysis, and some other core theoretical components of what we now consider to be solid engineering work were being applied long before the theoretical basis behind them was solidly proven to be correct from a pure mathematical standpoint.
Re:We don't understand it but we can do it (Score:3, Interesting)
The key idea here is that emergent algorithms (which is what these sort of things are) are unpredictable. It is one thing to understand the methodology, another to grok the full picture.
In any sort of complex input space, you cannot test all possible input permutations and so cannot guarantee that these algorithms wont go ahead and output the worst possible thing from time to time.
In some cases we can get away with emergent algorithms because the worst possible output isnt going to kill anybody (for example, using an ANTS algorithm for network routing is OK because the worst case is the already frequent phenomenon of a lost packet, and we are using the ANTS algorithm specifically because it does better routined and thus, in general loses fewer packets)
Re:We don't understand it but we can do it (Score:3, Interesting)
Re:We don't understand it but we can do it (Score:3, Interesting)
Here is my falsifiable hypothesis, I'll test it experimentally and see if successful results can be reproduced independently.
Theory comes a LONG ways down the road.