Tech_News

1. a node (called the root node) and
2. left and right sub-trees.
   Both the sub-trees are themselves binary trees.

1. the keys of all the nodes in the left sub-tree are less than that of the root,
2. the keys of all the nodes in the right sub-tree are greater than that of the root,
3. the left and right sub-trees are themselves ordered binary trees.

We might forgive a high-tech billionaire for making premature claims about the capabilities of autonomous vehicles. After all, keeping one between the relevant navigational beacons is piece of cake for special purpose algorithms running on digital hardware. The trouble is that those last few percentage points of occupant or bystander safety seem to get them every time. Many technology fans, including DARPA, think that brain-like neuromorphic chips that bleep digital spikes at each other could provide the answer — if only someone knew what exactly should generate the spikes they use. The latest brainstorm, before turning over the keys of life to these chips, is to put them into autonomous quadcopters and see what happens.
Obviously these neuromorphic chips are not just stuck onto the drones, but rather integrated somehow within the sensor-to-motor pipeline. Generally this means that the analog data from miscellaneous ultrasonic and optical sensors gets triaged in some fashion into neuro-digital spikes, which are propagated to higher level neurons
In the absence of the full physical vetting process (i.e. biology) used by real neurons, some kind of adaption algorithm is expediently imposed on individual neurons of the network. For some neurons, it could be as simple as something like this: Detection of big or noisy thing = spike fast, plus add some little twist to the network in lieu of real growing connections.

The quadcopter itself is custom built by AeroVironment, a unique company founded by Paul MacCready, with some funding from DARPA’s neuromorphic SyNAPSE project. Incidentally, the late MacCready became known internationally as the father of human-powered flight after building both the Gossamer Condor and Gossamer Albatross pedal craft. Not long ago, Bryan Adams, who first piloted the Albatross in a historic flight across the English Channel gave us a few insightful comments in one of our posts about human-powered flight. With its latest creation, AeroVironment has basically built a flying chip that continues its tradition of extreme lightness. At only 93 grams, 18 of those precious grams are the neuromorphic control chip.
The chip itself (shown in the center of the drone at the top of the story) is the product of an even more historic and iconic company, HRL Laboratories. Today Howard Hughes Research Labs is jointly owned by GM and Boeing. Having grown bored with trying to get huge wooden planes airborne (the Spruce Goose) or getting Ruby lasers to lase, HRL now dabbles in AI research — and apparently, neuromorphics. On a budget of just 50 milliwatts of power, its 576-neuron chip can recognize its surroundings and report to the drone whether or not they are in a familiar environment. As more details about this chip emerge, it would be interesting to compare its neural powers with more vision-specific drone control chips like for example, the neuromorphic bug eyecam.
 
A neuromorphic chip. (Not the one that powers the quadcopter.)
Not surprisingly, a few neuroscientists are anxious to get their hands on this chip. One thing they might like to try is to see if its network develops prosopagnosia (selective inability to recognize faces) when they cut its power down to 25 milliwatts. An even more telling sign that the chip is mimicking real brain function would be if the researchers could impart an intuitive emotional feel for different rooms or locations to the drone. The potential ability of the drone to feel different levels of safety or affinity for different places might be reflected in detection of the Capgras delusion when its emotion neurons were deactivated. In humans, this particular syndrome leaves the ability to detect faces intact, but impairs their ability to assign emotional content to the faces of those close to them. The curious result is that they claim the person is an impostor, a doppelganger of the real person who must be somewhere else.
One trend to observe in the control systems now being fleshed out for autonomous vehicles is the blooming demand for real time processors with advanced interrupt handling. As algorithms or threads compete for compute cycles and priority level either within a single processor or across several for control of the vehicle, conflicts will increasingly loom. At the extreme, the bulk of the effort in traditional computing systems then just becomes suspending threads and writing their incidentals to memory while higher priority events attempt to hold forth. In many ways, the neuro-digital spike itself can be thought of as an interrupt — all interrupt computing if you will. When the conversion from spike to bits, or from bits to spike becomes sufficiently muddled, then algorithm-free spikes must be both everything and nothing. In other words, not only are spikes a Morse code with that is all dots and no dashes, it is also a Morse code entirely devoid of any symbolic translation.
Practical neural chips are not quite ready to shed the algorithm entirely. The most important algorithm that awaits is probably not one that is network-wide, but rather one that simply dictates when a spike in an individual neuron should release a bit of transmitter. When that is in hand, (and when these chips even have any transmitter function built into them), then perhaps getting practical results out of 576-neuron network will be more straightforward.

source : http://www.extremetech.com/extreme/193532-darpas-new-autonomous-quadcopter-is-powered-by-a-brain-like-neuromorphic-chip