| CPC G06F 9/5027 (2013.01) [G06N 20/00 (2019.01)] | 16 Claims |
|
1. A pattern recognition system, comprising:
one or more processors;
memory communicatively coupled to the one or more processors, the memory storing instructions configured to, when executed by the one or more processors, implement a set of fractal cognitive computing nodes (FCNs);
wherein the set of FCNs is configured into a hierarchical network, the set of FCNs having child FCNs at lower levels of the hierarchical network coupled to parent FCNs at higher levels of the hierarchical network, each lower level of the hierarchical network having a respective higher resolution than each higher level of the hierarchical network;
wherein each child FCN is configured to:
provide a first respective set of tuples to a respective parent FCN connected thereto, each tuple in the first respective set of tuples comprising a first respective key and a first respective value, the first respective key being a first respective identifier of the tuple and the first respective value being a first respective magnitude of activation of the first respective identifier;
receive a second respective set of tuples from the respective parent FCN connected thereto or a respective input connection, each tuple in the second respective set of tuples comprising a second respective key and a second respective value, the second respective key being a second respective identifier of the tuple and the second respective value being a second respective magnitude of activation of the second respective identifier, the respective input connection providing data to be analyzed by the pattern recognition system;
wherein each parent FCN is configured to receive the first respective set of tuples from and provide the second respective set of tuples to respective child FCNs connected thereto; wherein each FCN is configured to:
determine, using the one or more processors, a respective similarity measure between each stored tuple of a set of tuples in the memory and a respective set of input tuples, the respective set of input tuples comprising the first respective set of tuples from a respective child FCN or the respective input connection, the second respective set of tuples from respective parent FCN, or both;
compare, using the one or more processors, the respective similarity measure to a spatial attention (SA) parameter;
in response to the respective similarity measure being lower than the SA parameter, add the stored tuple in the memory that corresponds to the respective similarity measure to a respective output set;
in response to each of the respective similarity measures being greater than the SA parameter, add the respective set of input tuples to the respective output set and a new stored tuple in the memory;
assign the respective output set as the first respective set of tuples, the second respective set of tuples, or both for the FCN;
determine whether a self-replication criterion is met based on content of the stored set of tuples and based on semantic or relationships between tuples; and
in response to determining that the self-replication criterion is met, replicate itself into one or more respective child FCNs at a respective lower level of the hierarchical network, wherein a first FCN replicates itself into a first new child FCN, the replicating comprising:
allocating first processing resources from the one or more processors to the first new child FCN;
allocating first memory resources from the memory to the first new child FCN;
increasing the SA parameter of the first new child FCN; and
connecting the first new child FCN into the hierarchical network according to first one or more keys of first one or more tuples in the first FCN,
wherein the first new child FCN processes tuples at a first sub-domain of a domain of the first FCN; and
wherein the pattern recognition system is configured to output the respective output sets for FCNs at a highest level of the hierarchical network.
|