Machine learning apparatus and method for machine learning by using tensor including plurality of axes as input
A non-transitory computer-readable recording medium having stored therein a machine learning program executable by one or more computers, the machine learning program including an instruction for generating a tensor comprising a first axis, a second axis, and a third axis, the first axis and the second axis representing relationships of a plurality of nodes included in graph information including data representing attributes of the plurality of nodes in the hierarchical structure, the third axis representing separately first data included in a first layer of the hierarchical structure and second data included in a second layer of the hierarchical structure, and an instruction for training a machine learning model by using the tensor as an input.
1. A non-transitory computer-readable recording medium having stored therein a machine learning program executable by one or more computers, the machine learning program comprising:
an instruction for collecting, from a memory, graph information including a plurality of nodes and data representing attributes of the plurality of nodes, the attributes being a hierarchical structure,
an instruction for generating a tensor being three dimensional structured data and comprising a first axis, a second axis, and a third axis, the first axis and the second axis representing relationships of the plurality of nodes included in the graph information, the third axis representing separately first data included in a first layer of the hierarchical structure and second data included in a second layer of the hierarchical structure, and
an instruction for training a machine learning model by using the tensor being the three dimensional structured data as an input, wherein
the generating of the tensor comprises generating the third axis representing the attributes by synthesizing, for each combination of a value of the first data and a value of the second data, the first data and the second data such that the first data and the second data according to the combination are distinguished from each other, and
the training comprises inputting a core tensor obtained by performing tensor decomposition of the tensor into the machine learning model, updating a parameter of a neural network of the model, and updating a parameter of the tensor decomposition.
2. The non-transitory computer-readable recording medium according to claim 1 , wherein
the generating of the tensor comprises generating the third axis by synthesizing, for each combination of nodes selected from the plurality of nodes, the first data and the second data related to one node of the combination such that the first data in the third axis is distinguished from the second data in the third axis.
3. The non-transitory computer-readable recording medium according to claim 2 , wherein
the generating of the tensor comprises performing an arithmetic operation, by using a vectorization operator, on information of a second slice of the second axis extracted from a first slice extracted from the first axis.
4. The non-transitory computer-readable recording medium according to claim 1 , the machine learning program further comprising an instruction for outputting a result of determination obtained by inputting a tensor including a first axis, a second axis, and a third axis that are generated, based on other graph information, into the machine learning model after the training.
5. The non-transitory computer-readable recording medium according to claim 1 , wherein
the first layer is an upper layer compared with the second layer.
6. A machine learning apparatus comprising:
a memory that stores graph information including a plurality of nodes and data representing attributes of the plurality of nodes, the attributes being a hierarchical structure; and
a processor coupled to the memory, the processor being configured to:
perform collection of the graph information from the memory,
perform generation of a tensor being three dimensional structured data and comprising a first axis, a second axis, and a third axis, the first axis and the second axis representing relationships of the plurality of nodes included in the graph information, the third axis representing separately first data included in a first layer of the hierarchical structure and second data included in a second layer of the hierarchical structure, and
perform training of a machine learning model by using the tensor being the three dimensional structured data as an input, wherein
the generation of the tensor comprises generating the third axis representing the attributes by synthesizing, for each combination of a value of the first data and a value of the second data, the first data and the second data such that the first data and the second data according to the combination are distinguished from each other, and
the training comprises inputting a core tensor obtained by performing tensor decomposition of the tensor into the machine learning model, updating a parameter of a neural network of the model, and updating a parameter of the tensor decomposition.
7. The machine learning apparatus according to claim 6 , wherein
the generation includes generating the third axis by synthesizing, for each combination of nodes selected from the plurality of nodes, the first data and the second data related to one node of the combination such that the first data in the third axis is distinguished from the second data in the third axis.
8. The machine learning apparatus according to claim 7 , wherein
the generation includes performing an arithmetic operation, by using a vectorization operator, on information of a second slice of the second axis extracted from a first slice extracted from the first axis.
9. The machine learning apparatus according to claim 6 , wherein the processor is further configured to output a result of determination obtained by inputting a tensor including a first axis, a second axis, and a third axis that are generated, based on other graph information, into the machine learning model after the training.
10. The machine learning apparatus according to claim 6 , wherein:
the first layer is an upper layer compared with the second layer.
11. A computer-implemented method for machine learning comprising:
performing collection, from a memory, of graph information including a plurality of nodes and data representing attributes of the plurality of nodes, the attributes being a hierarchical structure,
performing generation of a tensor being three dimensional structured data and comprising a first axis, a second axis, and a third axis, the first axis and the second axis representing relationships of the plurality of nodes included in the graph information, the third axis representing separately first data included in a first layer of the hierarchical structure and second data included in a second layer of the hierarchical structure, and
performing training of a machine learning model by using the tensor being the three dimensional structured data as an input, wherein
the generation of the tensor comprises generating the third axis representing the attributes by synthesizing, for each combination of a value of the first data and a value of the second data, the first data and the second data such that the first data and the second data according to the combination are distinguished from each other, and
the training comprises inputting a core tensor obtained by performing tensor decomposition of the tensor into the machine learning model, updating a parameter of a neural network of the model, and updating a parameter of the tensor decomposition.
12. The method according to claim 11 , wherein
the generation includes generating the third axis by synthesizing, for each combination of nodes selected from the plurality of nodes, the first data and the second data related to one node of the combination such that the first data in the third axis is distinguished from the second data in the third axis.
13. The method according to claim 12 , wherein
the generation includes performing an arithmetic operation, by using a vectorization operator, on information of a second slice of the second axis extracted from a first slice extracted from the first axis.
14. The method according to claim 11 , further comprising outputting a result of determination obtained by inputting a tensor including a first axis, a second axis, and a third axis that are generated, based on other graph information, into the machine learning model after the training.
15. The method according to claim 11 , wherein:
the first layer is an upper layer compared with the second layer.