Self-organizing data driven learning hardware with local interconnections
First Claim
Patent Images
1. A self-organizing apparatus, comprising:
- a plurality of artificial neurons arranged in multiple layers;
a plurality of initial interconnections interconnecting the plurality of artificial neurons to form an initial configuration of a self-organizing learning array in which less than total interconnection is undertaken between adjacent layers;
one or more learning array input nodes, each learning array input node connected to at least one artificial neuron; and
the plurality of artificial neurons including a first self-organizing artificial neuron comprising;
an entropy-based evaluator determining entropy of an input data space associated with the first self-organizing artificial neuron, wherein at least one of i) selection of initial interconnections to retain as neuron input connections to the first self-organizing artificial neuron and ii) selection of a transformation function to be applied by the first self-organizing artificial neuron is based at least in part on the determined entropy;
wherein the plurality of artificial neurons includes one or more self-organizing artificial neurons each self-organizing artificial neuron independently determines the initial interconnections associated therewith to retain as neuron input connections and the initial interconnections associated therewith to release in response to training data being applied to the one or more learning array input nodes, the retained interconnections forming a trained configuration of the self-organizing learning array;
wherein the first self-organizing artificial neuron determines the entropy of the input data space in response to training data being applied to the one or more learning array input nodes.
1 Assignment
0 Petitions
Accused Products
Abstract
A method for organizing processors to perform artificial neural network tasks is provided. The method provides a computer executable methodology for organizing processors in a self-organizing, data driven, learning hardware with local interconnections. A training data is processed substantially in parallel by the locally interconnected processors. The local processors determine local interconnections between the processors based on the training data. The local processors then determine, substantially in parallel, transformation functions and/or entropy based thresholds for the processors based on the training data.
-
Citations
44 Claims
-
1. A self-organizing apparatus, comprising:
-
a plurality of artificial neurons arranged in multiple layers; a plurality of initial interconnections interconnecting the plurality of artificial neurons to form an initial configuration of a self-organizing learning array in which less than total interconnection is undertaken between adjacent layers; one or more learning array input nodes, each learning array input node connected to at least one artificial neuron; and the plurality of artificial neurons including a first self-organizing artificial neuron comprising;
an entropy-based evaluator determining entropy of an input data space associated with the first self-organizing artificial neuron, wherein at least one of i) selection of initial interconnections to retain as neuron input connections to the first self-organizing artificial neuron and ii) selection of a transformation function to be applied by the first self-organizing artificial neuron is based at least in part on the determined entropy;wherein the plurality of artificial neurons includes one or more self-organizing artificial neurons each self-organizing artificial neuron independently determines the initial interconnections associated therewith to retain as neuron input connections and the initial interconnections associated therewith to release in response to training data being applied to the one or more learning array input nodes, the retained interconnections forming a trained configuration of the self-organizing learning array; wherein the first self-organizing artificial neuron determines the entropy of the input data space in response to training data being applied to the one or more learning array input nodes. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
-
-
16. A self-organizing artificial neuron, comprising:
-
one or more neuron data inputs adapted to receive corresponding input data signals; an input data multiplexer selectively connecting and disconnecting one or more of the input data signals; a neuronal processor selectively applying a transformation function to one or more connected input data signals, thus contributing to partitioning of an input data space associated with the self-organizing artificial neuron; and an entropy-based evaluator determining entropy of the input data space associated with the self-organizing artificial neuron, wherein at least one of i) selection of the input data signals for connection and disconnection and ii) selection of the transformation function to be applied by the self-organizing artificial neuron is based at least in part on the determined entropy; wherein the self-organizing artificial neuron selects the input data signals for, connection and disconnection in response to training data being operatively communicated to the self-organizing artificial neuron; wherein the self-organizing artificial neuron determines the transformation function to be applied in response to training data being operatively communicated to the self-organizing artificial neuron; wherein the self-organizing artificial neuron determines the entropy of the input data space in response to training data being operatively communicated to the self-organizing artificial neuron. - View Dependent Claims (17, 18)
-
-
19. A self-organizing artificial neuron, comprising:
-
one or more neuron data inputs adapted to receive corresponding input data signals; an input data multiplexer selectively connecting and disconnecting one or more of the input data signals; and an entropy-based evaluator determining entropy of an input data space associated with the self-organizing artificial neuron, wherein at least one of i) selection of the input data signals for connection and disconnection and ii) selection of a transformation function to be applied by the self-organizing artificial neuron is based at least in part on the determined entropy; wherein the self-organizing artificial neuron selects the input data signals for connection and disconnection in response to training data being operatively communicated to the self-organizing artificial neuron; wherein the self-organizing artificial neuron determines the entropy of the input data space in response to training data being operatively communicated to the self-organizing artificial neuron. - View Dependent Claims (20, 21)
-
-
22. A self-organizing artificial neuron, comprising:
-
one or more neuron data inputs adapted to receive corresponding input data signals; a neuronal processor selectively applying a transformation function to one or more input data signals, thus contributing to partitioning of an input data space associated with the self-organizing artificial neuron; and an entropy-based evaluator determining entropy of the input data space associated with the self-organizing artificial neuron, wherein at least one of i) selection of the input data signals for connection and disconnection at the self-organizing artificial neuron and ii) selection of the transformation function to be applied by the self-organizing artificial neuron is based at least in part on the determined entropy; wherein the self-organizing artificial neuron determines the transformation function to be applied in response to training data being operatively communicated to the self-organizing artificial neuron; wherein the self-organizing artificial neuron determines the entropy of the input data space in response to training data being operatively communicated to the self-organizing artificial neuron. - View Dependent Claims (23, 24)
-
-
25. A method for training a self-organizing apparatus, comprising:
-
a) arranging a plurality of artificial neurons in multiple layers, the plurality of artificial neurons including one or more self-organizing artificial neuron; b) interconnecting the plurality of artificial neurons with a plurality of initial interconnections to form an initial configuration of a self-organizing learning array, where less than total interconnection is undertaken between adjacent layers; c) connecting each of one or more learning array input nodes to at least one artificial neuron; d) applying training data to the one or more learning array input nodes; e) at each self-organizing artificial neurons, independently determining initial interconnections associated therewith to retain as neuron input connections and initial interconnections associated therewith to release in response to the training data applied in d); f) forming a trained configuration of the self-organizing learning array based at least in part on the retained interconnnections; and g) at a first self-organizing artificial neuron, independently determining entropy of an input data space associated with the first self-organizing artificial neuron based at least in part on the training data applied in d), wherein at least one of i) selecting initial interconnections to retain in e) and ii) selecting a transformation function to be applied by the first self-organizing artificial neuron is based at least in part on the determined entropy. - View Dependent Claims (26, 27, 28, 29, 30, 31, 32, 33, 34, 35)
-
-
36. A method for training a self-organizing artificial neuron, comprising:
-
a) receiving one or more input data signals at corresponding neuron data inputs of the self-organizing artificial neuron; b) at the self-organizing artificial neuron, independently and selectively connecting and disconnecting one or more of the input data signals, selection of the input data signals for connecting and disconnecting being based at least in part on training data operatively communicated to the self-organizing artificial neuron; c) at the self-organizing artificial neuron, independently and selectively applying a transformation function to one or more connected input data signals to at least partially partition an input data space associated with the self-organizing artificial neuron, selection of the transformation function to be applied being based at least in part on training data operatively communicated to the self-organizing artificial neuron; and d) at the self-organizing artificial neuron, independently determining entropy of the input data space associated with the self-organizing artificial neuron based at least in part on training data operatively communicated to the self-organizing artificial neuron, wherein at least one of i) selecting the input data signals for connecting and disconnecting in b) and ii) selecting the transformation function to be applied in c) is based at least in part on the determined entropy. - View Dependent Claims (37, 38)
-
-
39. A method for training a self-organizing artificial neuron, comprising:
-
a) receiving one or more input data signals at corresponding neuron data inputs of the self-organizing artificial neuron; b) at the self-organizing artificial neuron, independently and selectively connecting and disconnecting one or more of the input data signals, selection of the input data signals for connecting and disconnecting being based at least in part on training data operatively communicated to the self-organizing artificial neuron; and c) at the self-organizing artificial neuron, independently determining entropy of the input data space associated with the self-organizing artificial neuron based at least in part on training data operatively communicated to the self-organizing artificial neuron, wherein at least one of i) selecting the input data signals for connecting and disconnecting in b) and ii) selecting a transformation function to be applied by the self-organizing artificial neuron is based at least in part on the determined entropy. - View Dependent Claims (40, 41)
-
-
42. A method for training a self-organizing artificial neuron, comprising:
-
a) receiving one or more input data signals at corresponding neuron data inputs of the self-organizing artificial neuron; b) at the self-organizing artificial neuron, independently and selectively applying a transformation function to one or more input data signals to at least partially partition an input data space associated with the self-organizing artificial neuron, selection of the transformation function to be applied being based at least in part on training data operatively communicated to the self-organizing artificial neuron; and c) at the self-organizing artificial neuron, independently determining entropy of the input data space associated with the self-organizing artificial neuron based at least in part on training data operatively communicated to the self-organizing artificial neuron, wherein at least one of i) selecting input data signals for connecting and disconnecting at the self-organizing artificial neuron and ii) selecting the transformation function to be applied in b) is based at least in part on the determined entropy. - View Dependent Claims (43, 44)
-
Specification