HOME OF THE CREATIVITY MACHINE
The Big Bang of Machine Intelligence!
The IEI Blog is now open for discussion of new business ventures and opportunities!
Summary - IEI
and its newly formed subsidiary, Vigillect, has worked with Transportation
Security Administration (TSA) to develop truly clever and autonomous airport
security systems. The unique advantage of these brilliant sensor networks is
that they are constantly writing their own computer algorithms to detect the
subtlest deviations from normal airport operations. This aspect is especially
important when one considers the ever changing nature of the airport
environment, as well as the cost and hassle of engaging humans to constantly
modify system rules.
to these efforts was IEI’s coveted ability to coerce immense neural networks
to knit themselves into brain-like cascades that can detect and classify
anomalous objects and scenarios on the airport grounds. One very impressive
accomplishment under our Phase I research under TSA was the detection of unusual
scenarios at the passenger drop off areas that included identification of over
parked vehicles as well as suspicious pedestrian milling behaviors. In similar
fashion, our engineers were able to isolate and train especially large neural
nets to recognize authorized vehicles and personnel. Introducing
weight-in-motion readings from arriving vehicles we were able to detect
suspiciously overloaded cars and trucks. Using bi-static radar returns from the
parking lot we could also pinpoint unusual pedestrian and vehicular scenarios
taking place therein.
in Phase II of this TSA sponsored project, IEI was able to develop a similar
sensor network for airport ramps and runways that could autonomously function
round the clock, constantly forming combined spatial and temporal models of
airport normalcy. While learning to ignore routine activity, the system could
readily detect and alarm on unscheduled aircraft arrivals, as well as
unauthorized personnel and stray animals on the runways and tarmac. Similarly,
unauthorized vehicles and personnel could be identified at key entrance points.
Finally, all neural nets from all nodes on the airport perimeter fed a master
neural network cascade that constantly learned overall airport routine and could
intelligently alert airport administrators to suspicious activities with minimal
false alarm rates.
key contributing factors to our successes on this and related military projects
has been our unique ability to recreate many of the brain’s cognitive
functions using inventive, self-connecting cascades of artificial neural
networks, so-called “Supernets,”
that have invented whole new methodologies and capabilities for performing a
wide range of security-related functions.
instance, using our extremely efficient neural networks devised not by humans,
but by our inventive neural architectures, we are able to process all million or
so bytes from each camera frame of a video stream, while preserving a 20-30 fps
frame rate as the underlying networks learn to identify objects, scenarios, and
deviations from status quo activity. Such machine-conceived neural nets are
capable of developing an environmental normalcy model that is capable of
evolving to keep pace with ever changing lighting conditions. Simultaneously,
such nets are capable of highlighting, in very brain-like fashion, what
doesn’t intrinsically belong to the scene in manner that is vastly superior to
rather simplistic frame subtraction techniques and motion detection algorithms
currently in use.
our creative AI has invented an exceptional methodology for the isolation of
newly arriving objects on the scene in a manner that is unusually resistant to
environmental lighting fluctuations. This accomplishment has been the long
sought after “holy grail” of camera-based anomaly detection efforts.
Thereafter, the freshly extracted object or scenario can be passed to other
self-organizing neural cascades for classification.
to a robust classification is a neural architecture and methodology, first
conceived by our inventive neural nets, that is called a “Group Membership
Filter” or “GMF.”
Such GMFs self-organize themselves to recognize a given object or scenario over
a wide range of sensor perspectives. This approach is vastly different in
philosophy from the standard neural network approach wherein certain attributes
are calculated from imagery and then compared against a previously generated
database in a process that has become known as “registration.” In a sense,
our GMFs become expert at identifying a thing or an action based upon multiple
presentations of a given genre to it, without the need for training upon
counter-examples. Furthermore, once identifying a target, they may lock on to it
and automatically track it, constantly learning all of the alternative forms it
may take as a result of changing orientation or illumination.
the end, the extremely fast and large artificial neural networks we call
“Self-Training Artificial Neural Network Objects” (STANNO) were able to
interconnect via TCP/IP to create vast, brain-like governing layer that could
now develop airport-wide normalcy models, as well as resolve ambiguous sensor
inputs from the LAN’s distributed nodes. At the core of this immense neural
architecture was a recently completed graphical programming library that allows
us to drag and drop our STANNOs into a wiring diagram so that they could
interconnect with themselves and supplied device interfaces to form these
security-minded cognitive structures. We were thus able to rapidly prototype and
refine candidate systems in the course of this TSA activity, and in so doing
developed a methodology for tailoring our contemplative artificial intelligence
systems to the suite of available sensors and actuators.
ability of our systems to deal with and adapt to the unexpected, drawing upon
our patented “Creativity
Machine” neural architectures to cleverly devise the most effective course
of action to a potentially harmful scenario. Such systems may either be mentored
in advance by humans or bootstrap their competencies via our latest generation
of self-correcting Creativity Machines. As is the case with all IEI ventures,
the dependency upon human beings has been reduced, leading to lower costs,
tireless vigilance, and much broader, human-like capabilities.
© 1997-2013, Imagination Engines, Inc. | Creativity Machine®, Imagination Engines®, and DataBots® are registered trademarks of Imagination Engines, Inc.
1550 Wall Street, Ste. 300, St. Charles, MO 63303 • (636) 724-9000