Analysis device, method, and program

A computer-implemented method for analyzing aspects of a neural network, the method comprising:

• receiving the neural network, wherein the neural network is previously learnt, and wherein the neural network includes input data, a plurality of units, and output data;
• determining a first strength of a first relationship between a first dimension of the input data and a unit of the plurality of units of the neural network;
• determining a second strength of a second relationship between the unit and a second dimension of the output data of the neural network;
• determining, based on the first strength of the first relationship and the second strength of the second relationship, at least one of:
• a third relationship between a prescribed number of types of roles and the unit,
• a fourth relationship between the prescribed number of types of roles and the first dimension of the input data, and
• a fifth relationship between the prescribed number of types of roles and the second dimension of the output data;
• and
• providing, based at least in part on the determined third relationship, the determined fourth relationship, and the determined fifth relationship, interactive output.

A system for analyzing aspects of a neural network, the system comprises:

• a processor;
• and
• a memory storing computer-executable instructions that when executed by the processor cause the system to:
• receive the neural network, wherein the neural network is previously learnt, and wherein the neural network includes input data, a plurality of units, and output data;
• determine a first strength of a first relationship between a first dimension of the input data and a unit of the plurality of units of the neural network;
• determine a second strength of a second relationship between the unit and a second dimension of the output data of the neural network;
• determine, based on the first strength of the first relationship and the second strength of the second relationship, at least one of:
• a third relationship between a prescribed number of types of roles and the unit,
• a fourth relationship between the prescribed number of types of roles and the first dimension of the input data, and
• a fifth relationship between the prescribed number of types of roles and the second dimension of the output data;
• and
• provide, based at least in part on the determined third relationship, the determined fourth relationship, and the determined fifth relationship, interactive output.

A computer-readable non-transitory recording medium storing computer-executable instructions that when executed by a processor cause a computer system to:

• receive the neural network, wherein the neural network is previously learnt, and wherein the neural network includes input data, a plurality of units, and output data;
• determine a first strength of a first relationship between a first dimension of the input data and a unit of the plurality of units of the neural network;
• and
• determine a second strength of a second relationship between the unit and a second dimension of the output data of the neural network;
• determine, based on the first strength of the first relationship and the second strength of the second relationship, at least one of:
• a third relationship between a prescribed number of types of roles and the unit,
• a fourth relationship between the prescribed number of types of roles and the first dimension of the input data, and
• a fifth relationship between the prescribed number of types of roles and the second dimension of the output data;
• and
• provide, based at least in part on the determined third relationship, the determined fourth relationship, and the determined fifth relationship, interactive output.