| CPC G16H 50/50 (2018.01) [G06N 20/00 (2019.01); G16H 30/20 (2018.01); G06F 18/213 (2023.01); G06F 18/22 (2023.01)] | 5 Claims |
|
1. A registration apparatus that reduces a calculation period that occurs when a machine learning network is updated to include a new node, the registration apparatus comprising:
a memory;
a communication interface that is communicatively coupled to the machine learning network, wherein the machine learning network includes a set of first nodes and first edges, the first nodes each representing a first feature vector including a plurality of elements, the first edges each coupling two first nodes representing two first feature vectors to each other based on two first feature vectors;
a processor that is communicatively coupled to the memory and the communication interface,
wherein the processor is configured to:
obtain a second feature vector,
obtain first correlation data, which represents correlation between each pair of first feature vectors in the machine learning network,
determine whether a first similar group of specific third feature vectors exists in the set of third feature vectors, the specific third feature vectors satisfying a predetermined similarity condition,
obtain correlation data, which represents correlation between each pair of the first feature vectors in the machine learning network,
calculate a second correlation data indicating correlation between the second feature vector and each of the third feature vectors and correlation between each pair of the third feature vectors,
extract, when the first similar group is determined not to exist, from the first correlation data, third correlation data having a same size as a size of the second correlation data calculated,
calculate a similarity between the second correlation data and the third correlation data extracted,
identify, from the first correlation data, a second similar group of specific first feature vectors being a calculation source of the third correlation data based on the similarity calculated,
determine, as a registration destination for the new node, a position in the machine learning network based on the specific first feature vectors included in the second similar group and the specific third feature vectors included in the first similar group,
register the new node to the registration destination, wherein the new node represents the second feature vector based on a similarity relationship among the third feature vectors in a set of the third feature vectors included in the set of first feature vectors, and
couple the new node and a third node representing the third feature vector to each other with a second edge, wherein a number of the third feature vectors is smaller than a number of the first feature vectors.
|