Classification apparatus, classification method, and program

A classification method performed by a computer, the method comprising:

• an encoding process that includes:
• electrically receiving sensing data as input data from a sensor,
• performing an element classification process that extracts a feature of the input data and outputs classification information, based on an element classification model implemented by a neural network and stored in a first storage unit, wherein the classification information X is represented by a pair of first and second information elements X=(x1,x2), where the first information element x1 and the second information element x2 indicate a location of a peak in a waveform of the input data and a value of the peak, respectively;
• an element classification model learning process that includes
• generating element classification model based on inputted element data ϕ and ψ and stores the element classification model in the first storage unit, wherein the element data includes a first set of locations of peaks of the input data and a second set of peak values corresponding to the peaks included in the first set;
• an integration process that includes
• receiving receives the classification information X=(x1,x2) and converts the classification information to a collation vector Z=(z1,z2), based on an element estimation model stored in a second storage unit generated by an element estimation model learning process, wherein the element estimation model includes a probability p(ϕ|x1=a) which is a conditional probability with which the element data is ϕ when x1=a is given and a probability p(ψ|x2=b) which is a conditional probability with which the element data is ψ when x2=b is given for the inputted classification information X=(x1,x2);
• a determination process that includes
• determining a group to which the collation vector Z=(z1,z2) belongs by collating the collation vector Z=(z1,z2) with a representative vector of an individual group stored as a semantic model in a third storage unit to output a group ID of the group as a classification result, and
• the element estimation model learning process generates the element estimation model based on the classification information and the element data to store the element estimation model in the second storage unit, wherein the element estimation model learning process learns the element estimation model by using, as input, the group ID outputted from the determination process.

A non-transitory computer-readable medium storing a program causing a computer to execute:

• an encoding process that includes:
• electrically receiving sensing data as input data from a sensor,
• performing an element classification process that extracts a feature of the input data and outputs classification information, based on an element classification model implemented by a neural network and stored in a first storage unit, wherein the classification information X is represented by a pair of first and second information elements X=(x1,x2), where the first information element x1 and the second information element x2 indicate a location of a peak in a waveform of the input data and a value of the peak, respectively;
• an element classification model learning process that includes
• generating the element classification model based on inputted element data ϕ and ψ and stores the element classification model in the first storage unit, wherein the element data includes a first set of locations of peaks of the input data and a second set of peak values corresponding to the peaks included in the first set;
• an integration process that includes
• receiving receives the classification information X=(x1,x2) and converts the classification information to a collation vector Z=(z1,z2), based on an element estimation model stored in a second storage unit generated by an element estimation model learning process, wherein the element estimation model includes a probability p(ϕ|x1=a) which is a conditional probability with which the element data is ψ when x1=a is given and a probability p(ψ|x2=b) which is a conditional probability with which the element data is when x2=b is given for the inputted classification information X=(x1,x2);
• a determination process that includes
• determining a group to which the collation vector Z=(z1,z2) belongs by collating the collation vector Z=(z1,z2) with a representative vector of an individual group stored as a semantic model in a third storage unit to output a group ID of the group as a classification result, and
• the element estimation model learning process generates the element estimation model based on the classification information and the element data to store the element estimation model in the second storage unit, wherein the element estimation model learning process learns the element estimation model by using, as input, the group ID outputted from the determination process.