Slashdot Mirror
DARPA Testing Numenta's Brain Tech
lousyd writes "CNN Money reports that DARPA and the National Geospatial-Intelligence Agency have given $4.9 million to Lockheed Martin to develop an image recognition system that will be used to scan satellite images and photographs for familiar objects. Called Object Recognition via Brain-Inspired Technology (ORBIT), the system will fuse commercial airborne EO and LIDAR sensor data into a three-dimensional, photo-realistic model of the landscape. The brains of the system, so to speak, will be Numenta's Hierarchical Temporal Memory technology, modeled on the technology growing inside human heads. The system is expected to increase image analysts' productivity by 100 times."
While loaded with buzzwords, this really involves nothing that's really new. The HTM is just a rehash of Adaptive Resonance Theory .ps.gz file).
And applications like this aren't exactly new (this link downloads a
Although it is certainly a major engineering challenge to get this type of classification to work over multiple modalities of data in any coherent way, as far as I can tell this project doesn't represent any breakthrough in approach or capability.
Urban Reasoning and Geospatial Exploitation Technology (URGENT)
The Urban Reasoning and Geospatial Exploitation Technology (URGENT) program is will develop a 3D urban object recognition and exploitation system that enables advanced mission planning and situation analysis capabilities for the warfighter operating in urban environments.
The recognition of targets in urban environments poses unique operational challenges for the warfighter. Historically, target recognition has focused on conventional military objects, with particular emphasis on military vehicles such as tanks and armored personnel carriers. In many cases, these threats exhibit unique signatures and are relatively geographically isolated from densely populated areas. The same cannot be said of today's asymmetric threats, which are embedded in urban areas, thereby forcing U.S. Forces to engage enemy combatants in cities with large civilian populations. Under these conditions, even the most common urban objects can have tactical significance: trash cans can contain improvised explosive devices, doors can conceal snipers, jersey barriers can block troop ingress, roof tops can become landing zones, and so on. Today's urban missions involve analyzing a multitude of urban objects in the area of regard. As military operations in urban regions have grown, the need to identify urban objects has become an important requirement for the military. Understanding the locations, shapes, and classifications of objects is needed for a broad range of pressing urban mission planning analytical queries (e.g., finding all roof top landing zones on three story buildings clear of vertical obstructions and verifying ingress routes with maximum cover for ground troops). In addition, it will enable automated time-sensitive situation analysis (e.g., alerting for vehicles found on a road shoulder after dark and estimating damage to a building exterior after an explosion) that will make a significant positive impact on urban operations.
Phase 1 of the URGENT program is developing techniques for the rapid exploitation of EO and LIDAR sensor data at the city scale to recognize urban objects down to the soldier scale. URGENT is applying image processing technology to geospatially registered 2D/3D data collected from airborne and terrestrial sources, yielding precise annotations for the objects in an urban area.
Phase 2 of the URGENT program will develop a 3D reasoning engine to query over object shapes, locations, and classifications for rapid urban mission planning, mission rehearsal, and situation analysis. Phase 3 will focus on the integration and transition of the URGENT system to the National Geospatial-Intelligence Agency (NGA).
How we know is more important than what we know.
The good news is that this is all math! There's no need to believe anything one way or another! Sorta exciting huh? You can go and examine all the ART algorithms (I linked wikipedia because it has the PDFs linked.. did you notice? But here's Grossberg's homepage, just in case), and you can go read about HTM. According to Hawkins, HTM has some magical, er I mean, proprietary, component that separates it from ART. I've seen Hawkins speak... in fact, I saw him speak at BU with Steve Grossberg in the audience. He amused the audience by showing a demo that was completely indistinguishable from an ART1 implementation that takes about half an hour to program, and most of the people present had done themselves.
He then failed to answer any substantive questions (including Steve asking him how his model differed from ART), referring us all to online videos of his lectures. I personally asked about how he could reconcile this article with his predictions.. which assume a cortical hierarchy based on 'distance' (in synapses) from primary sensory cortices, rather than examining the relative lamination of various cortices. I notice since then the wikipedia article "On Intelligence" has had its 'experimental prediction' claims toned down quite a bit.
As it happens in terms of books though, Grossberg has written several and has a ton of peer reviewed articles on this very subject. Hawkins to my knowledge doesn't have a single peer-reviewed article on HTM or anything related.
"He who appeals to authority when there is a difference of opinion works with his memory rather than his reason."
Please consider this post about Jeff Hawkins' history of navel-gazing idiocy in the field of neural networks.
You worship engineers? Why this one in any case? Even if he is a good engineer, that doesn't make him a good scientist (incidentally, he's not). Maybe you're not an expert in neural networks and are deferring to someone who is, at least plausibly, an expert. But you just illustrated that you're no better a thinker than the people you probably decry day-in, day-out; you're just one of their number who reads slashdot and gets an undeserved superiority complex. Did you even consider that "some guy on slashdot" provided a link to an established area of theory that is (if you investigated even a bit) AT LEAST similar to Hawkins' stuff?
Jeff Hawkins is not even noteworthy in the fields of neural networks or AI in general. He's got a lot of bluster and a lot of money though, and he's using that money to get the popular (in geek circles) press gain the "prestige" that can only be bought through useful, novel research in academia. He's not an expert, and he's probably a charlatan for claiming to be a cutting-edge researcher merely commercializing "his" (ahem) "new" (ahem) breakthroughs.
His disingenuous pretense is what gets me about him. Your unthinking deference to "what you read in some book" is what gets me. Please, if you're interested in neural nets (you seem to be), read some of the cornerstones in the field, read the academic literature, and stay away from Numenta and Hawkins.