Neural directors
AssigneeThe United States of America as represented by the Secretary of the Navy
InventorRoger L. Woodall
Life of this patent
Estimating the term…
Abstract
A neural director is provided which is a neural network constructed with weights that are determined a priori. The neural director receives an input vector X comprising “I” input components “Xi” and generates in response an output vector Y comprising “J” output components. The neural director has an input processing node layer which receives the input vector X and an output processing node layer which generates the output vector Y. The connections between the input and output processing node layers are a unique weighting set w(i,j) that contains an internal representation of a uniform spatial distribution of “J” unit vectors throughout a unit sphere of “I” dimensions. Thus the cosine value between any two adjacent unit vectors is a constant everywhere in the unit sphere.
Claims17
A neural director comprising:
- an input processing node layer having a first plurality “I” of input processing nodes;
- an output processing node layer having a second plurality “J” of output processing nodes;
- and
- connections from each of the input processing nodes to each of the output processing nodes, each one of the connections constructed with a weighting value corresponding to one element of a set of “J” unit vectors uniformly distributed in a spherical space of “I” dimensions wherein a cosine value between any two adjacent unit vectors of the set of unit vectors is a constant.
A method for determining weighting values of connections in a neural director having an input processing node layer with a first plurality “I” of input processing nodes-and an output processing node layer having a second plurality “J” of output processing nodes, the connections being from each of the input processing nodes to each of the output processing nodes, the method comprising the steps of:
- generating an array of “J” unit vectors uniformly distributed in a spherical space of “I” dimensions wherein a cosine value between any two adjacent unit vectors of the set of unit vectors is a constant;
- and
- assigning one component value of the array to each connection, such that the weighting value of a connection between the ith input processing node and the jth output processing node corresponds to the ith component value of the jth unit vector.
A method for determining weighting values of connections in a neural director having an input processing node layer with a first plurality “I” of input processing nodes and an output processing node layer having a second plurality “J” of output processing nodes, the connections being from each of the input processing nodes to each of the output processing nodes, the method comprising the steps of:
- placing an array of “J” random direction unit vectors in a unit spherical space of “I” dimensions, the placing step further comprising the steps of:
- randomly selecting one of the unit vectors;
- determining from among neighboring unit vectors, a neighboring unit vector nearest the selected unit vector;
- selecting an element of the selected unit vector;
- randomly altering a value of the element;
- determining if the altering results in movement of the selected unit vector in one of a direction away from the nearest neighboring unit vector and a direction not into a lesser distance towards at least one of the neighboring unit vectors not the nearest neighboring unit vector;
- and
- returning to the step of randomly selecting a unit vector with the altered unit vector normalized to a new unit vector when the altering results in movement;
- and
- assigning one element value of the array to each connection, such that the weighting value of a connection between an ith input processing node and a jth output processing node corresponds to an ith element value of a jth unit vector.
Dependencies are inferred from the claim text.