Method for determination of spatial target probability using a model of multisensory processing by the brain
First Claim
Patent Images
1. A method of determining spatial target probability comprising the steps of:
- acquiring at least two inputs from a location in a desired environment;
applying said inputs to a plurality of model units in a map corresponding to a plurality of locations in said environment;
approximating a conditional probability of a first target at each of said model units based on said at least two inputs;
finding a model unit from said plurality of model units with a highest conditional probability;
choosing a location in said environment corresponding to said model unit with said highest conditional probability as a location of a next target.
4 Assignments
0 Petitions
Accused Products
Abstract
A method of determining spatial target probability using a model of multisensory processing by the brain includes acquiring at least two inputs from a location in a desired environment where a first target is detected, and applying the inputs to a plurality of model units in a map corresponding to a plurality of locations in the environment. A posterior probability of the first target at each of the model units is approximated, and a model unit with a highest posterior probability is found. A location in the environment corresponding to the model unit with a highest posterior probability is chosen as the location of the next target.
119 Citations
40 Claims
-
1. A method of determining spatial target probability comprising the steps of:
-
acquiring at least two inputs from a location in a desired environment;
applying said inputs to a plurality of model units in a map corresponding to a plurality of locations in said environment;
approximating a conditional probability of a first target at each of said model units based on said at least two inputs;
finding a model unit from said plurality of model units with a highest conditional probability;
choosing a location in said environment corresponding to said model unit with said highest conditional probability as a location of a next target. - View Dependent Claims (2, 3, 4, 5, 6, 7)
-
-
8. A method of determining spatial target probability using a supervised learning algorithm in a model of a neural network having a plurality of input units, output units and hidden units connected between said input and output units, said method comprising the steps of:
-
training the model neural network to reduce an error between an actual response and a desired response of the neural network to predetermined inputs from a known location in a desired environment;
applying at least one input associated with a first target located in said desired environment;
finding an output unit from the plurality of output units with a highest desired value; and
choosing a location in said environment corresponding to said output unit with said highest desired value as a location of a next target. - View Dependent Claims (9, 10, 11, 12, 13)
positioning a training target at a random location in said desired environment;
applying at least one input associated with said training target to the model neural network to obtain said actual responses of the model neural network to said training target;
generating said desired responses of the model neural network to said training target;
finding differences between said actual and desired responses; and
using back-propagation to reduce said differences between said actual and desired responses.
-
-
12. The method as defined in claim 11, wherein said step of using back-propagation includes iteratively adjusting weights associated with the hidden units.
-
13. The method as defined in claim 10, wherein said sensory inputs include audio and video inputs.
-
14. An apparatus for automatically tracking a target in a desired environment, said system comprising:
-
at least one first sensor for receiving sensory inputs from the target;
a controller, based on said sensory inputs, for locating the target in the environment using a program modeling a neural network of a brain; and
at least one directional second sensor for turning to a location in the environment where the target has been located by said controller, wherein said model of said neural network includes a map having a plurality of model units corresponding to a plurality of locations in the environment for receiving information from said sensory inputs associated with the target located in the environment through a plurality of input units and connections between said input units and said model units. - View Dependent Claims (15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31)
said model neural network is trained to reduce an error between an actual response and a desired response of said model neural network to predetermined inputs from a known location in the environment; sensory inputs associated with the target located in the environment is applied to said plurality of inputs of said model neural network;
the model units with a highest desired value is found; and
a location in the environment corresponding to said model unit with said highest desired value is chosen as a location of a next target.
-
-
22. The apparatus as defined in claim 20 wherein said sensory inputs are audio and video signals.
-
23. The apparatus as defined in claim 21, wherein said model neural network is trained by,
positioning a training target at a random location in the predefined environment; -
applying sensory inputs associated with said training target to the model neural network to obtain said actual responses of the model neural network to said training target;
generating said desired responses of the model neural network to said training target;
finding differences between said actual and desired responses; and
using back-propagation to reduce said differences between said actual and desired responses.
-
-
24. The apparatus as defined in claim 14 wherein the target is located by an unsupervised adaptive algorithm in which,
said map is adjusted using a Kohonen mechanism to increase an amount of information from said sensory inputs that are transmitted to said map; - and
a strength of said sensory inputs associated with the target is modulated based on a correlation between activities of said map and predefined modulation units, and on anti-correlation between said predefined modulation units and said sensory inputs associated with said target.
- and
-
25. The apparatus as defined in claim 24 wherein said Kohonen mechanism adjusts weights associated with said connections between said input units and said model units such that each of said model units become specialized for receiving information indicating a predetermined location in said environment.
-
26. The apparatus as defined in claim 25 wherein said model units further become specialized for receiving a predetermined modality of said sensory inputs associated with said target.
-
27. The apparatus as defined in claim 26 wherein said modality includes at least audio and video inputs.
-
28. The apparatus as defined in claim 26 wherein said modulation units are predefined according to a modality of said sensory inputs associated with said target.
-
29. The apparatus as defined in claim 28 wherein said modulation units modulate said strength of said sensory inputs by multiplying weights associated with said sensory inputs.
-
30. The apparatus as defined in claim 28 wherein said modality includes at least audio and video inputs.
-
31. The apparatus as defined in claim 28 wherein at least one of said modulation units predefined by a first modality of said sensory inputs becomes active when said map receives information through at least said first modality form said sensory inputs, said at least one of said modulation units decreases modulation of said sensory inputs having first modality and increases modulation of said sensory inputs having modality other than said first modality.
-
32. A method of determining spatial target probability using an unsupervised adaptive algorithm in a model of a neural network, said method comprising the steps of:
-
organizing a map into a plurality of model units corresponding to a plurality of locations in a desired environment for receiving information from sensory inputs associated with a target located in said environment through a plurality of input units and connections between said input units and said model units;
adjusting said map to increase an amount of said information from said sensory inputs that are transmitted to said map using an unsupervised learning mechanism; and
modulating a strength of said sensory inputs associated with said target based on a correlation between activities of said map and predefined modulation units, and on anti-correlation between said predefined modulation units and said sensory inputs associated with said target. - View Dependent Claims (33, 34, 35, 36, 37, 38, 39, 40)
-
Specification