Information processing system, information processing method, and storage medium

1 . An information processing system, comprising:

at least one processor,

wherein the at least one processor is configured to:

acquire a numerical representation and a compounding ratio for each of a plurality of first polymers, wherein, for each of the plurality of first polymers, the numerical representation includes a numerical value indicating a chemical structure of a first polymer;

acquire a numerical representation for each of a plurality of second polymers, wherein, for each of the plurality of second polymers, the numerical representation includes a numerical value indicating a chemical structure of a second polymer;

generate a first trained machine learning model by executing and repeating a first machine learning process comprising:

inputting a first input vector into a first untrained machine learning model to calculate a first output vector;

calculating a first estimation result from the first output vector;

calculating a first difference between the first estimation result and a first ground truth; and

updating a first parameter of the first untrained machine learning model based on the first difference;

calculate a first feature vector of each of the plurality of first polymers and a second feature vector of each of the plurality of second polymers by inputting a plurality of the numerical representations corresponding to the plurality of first polymers and a plurality of the numerical representations corresponding to the plurality of second polymers into the first trained machine learning model;

generate a second trained machine learning model by executing and repeating a second machine learning process comprising:

inputting a second input vector into a second untrained machine learning model to calculate a second output vector;

calculating a second estimation result from the second output vector;

calculating a second difference between the second estimation result and a second ground truth; and

updating a second parameter of the second untrained machine learning model based on the second difference;

calculate a probability vector indicating a degree of association with each of the plurality of second polymers, for each of the plurality of first polymers, by inputting a plurality of the first feature vectors and a plurality of the second feature vectors into the second trained machine learning model, wherein the second trained machine learning model is one selected from a group consisting of an attention recurrent neural network (RNN) model and a multi-head attention model, each configured to calculate the probability vector while taking into account an influence of a combination of the plurality of the first feature vectors and the plurality of the second feature vectors;

calculate a composite feature vector indicating a degree of association with each of the plurality of second polymers, for a polymer alloy obtained by combining the plurality of first polymers, based on a plurality of the probability vectors and a plurality of the compounding ratios, wherein the at least one processor is further configured to:

cause a dimensionality of the probability vector to match a number of the plurality of second polymers; and

cause a structure of the composite feature vector to be identical to a structure of the probability vector;

calculate a predicted value of a characteristic of the polymer alloy by inputting the composite feature vector into a third trained machine learning model, wherein the predicted value indicates at least one of glass transition temperature and elastic modulus; and

cause a display device to display the predicted value.

2 . The information processing system according to claim 1 ,

wherein when the number of the plurality of first polymers is denoted by m, the number of the plurality of second polymers is denoted by n, each of the plurality of probability vectors is n-dimensional, the plurality of probability vectors are denoted by Vp 1 , Vp 2 , . . . , Vp m , the plurality of compounding ratios are denoted by r 1 , r 2 , . . . , r m , and the composite feature vector is denoted by Vc, the at least one processor is configured to calculate the composite feature vector by using following Equation (1),

V

c

=

r

1

×

V

⁢

p

1

+

r

2

×

V

⁢

p

2

+

…

+

r

m

×

V

⁢

p

m

(

1

)

3 . An information processing method executed by an information processing system including at least one processor, the method comprising:

acquiring a numerical representation and a compounding ratio for each of a plurality of first polymers, wherein, for each of the plurality of first polymers, the numerical representation includes a numerical value indicating a chemical structure of a first polymer;

acquiring a numerical representation for each of a plurality of second polymers, wherein, for each of the plurality of second polymers, the numerical representation includes a numerical value indicating a chemical structure of a second polymer;

generating a first trained machine learning model by executing and repeating a first machine learning process comprising:

inputting a first input vector into a first untrained machine learning model to calculate a first output vector;

calculating a first estimation result from the first output vector;

calculating a first difference between the first estimation result and a first ground truth; and

updating a first parameter of the first untrained machine learning model based on the first difference;

calculating a first feature vector of each of the plurality of first polymers and a second feature vector of each of the plurality of second polymers by inputting a plurality of the numerical representations corresponding to the plurality of first polymers and a plurality of the numerical representations corresponding to the plurality of second polymers into the first trained machine learning model;

generating a second trained machine learning model by executing and repeating a second machine learning process comprising:

inputting a second input vector into a second untrained machine learning model to calculate a second output vector;

calculating a second estimation result from the second output vector;

calculating a second difference between the second estimation result and a second ground truth; and

updating a second parameter of the second untrained machine learning model based on the second difference;

calculating a probability vector indicating a degree of association with each of the plurality of second polymers, for each of the plurality of first polymers, by inputting a plurality of the first feature vectors and a plurality of the second feature vectors into the second trained machine learning model, wherein the second trained machine learning model is one selected from a group consisting of an attention recurrent neural network (RNN) model and a multi-head attention model, each configured to calculate the probability vector while taking into account an influence of a combination of the plurality of the first feature vectors and the plurality of the second feature vectors;

calculating a composite feature vector indicating a degree of association with each of the plurality of second polymers, for a polymer alloy obtained by combining the plurality of first polymers, based on a plurality of the probability vectors and a plurality of the compounding ratios, wherein the at least one processor is configured to:

cause a dimensionality of the probability vector to match a number of the plurality of second polymers; and

cause a structure of the composite feature vector to be identical to a structure of the probability vector;

calculating a predicted value of a characteristic of the polymer alloy by inputting the composite feature vector into a third trained machine learning model, wherein the predicted value indicates at least one of glass transition temperature and elastic modulus; and

causing a display device to display the predicted value.

4 . A non-transitory computer-readable storage medium storing an information processing program causing a computer to execute:

acquiring a numerical representation and a compounding ratio for each of a plurality of first polymers, wherein, for each of the plurality of first polymers, the numerical representation includes a numerical value indicating a chemical structure of a first polymer;

acquiring a numerical representation for each of a plurality of second polymers, wherein, for each of the plurality of second polymers, the numerical representation includes a numerical value indicating a chemical structure of a second polymer;

generating a first trained machine learning model by executing and repeating a first machine learning process comprising:

inputting a first input vector into a first untrained machine learning model to calculate a first output vector;

calculating a first estimation result from the first output vector;

calculating a first difference between the first estimation result and a first ground truth; and

updating a first parameter of the first untrained machine learning model based on the first difference;

calculating a first feature vector of each of the plurality of first polymers and a second feature vector of each of the plurality of second polymers by inputting a plurality of the numerical representations corresponding to the plurality of first polymers and a plurality of the numerical representations corresponding to the plurality of second polymers into the first trained machine learning model;

generating a second trained machine learning model by executing and repeating a second machine learning process comprising:

inputting a second input vector into a second untrained machine learning model to calculate a second output vector;

calculating a second estimation result from the second output vector;

calculating a second difference between the second estimation result and a second ground truth; and

updating a second parameter of the second untrained machine learning model based on the second difference;

calculating a probability vector indicating a degree of association with each of the plurality of second polymers, for each of the plurality of first polymers, by inputting a plurality of the first feature vectors and a plurality of the second feature vectors into the second trained machine learning model, wherein the second trained machine learning model is one selected from a group consisting of an attention recurrent neural network (RNN) model and a multi-head attention model, each configured to calculate the probability vector while taking into account an influence of a combination of the plurality of the first feature vectors and the plurality of the second feature vectors;

calculating a composite feature vector indicating a degree of association with each of the plurality of second polymers, for a polymer alloy obtained by combining the plurality of first polymers, based on a plurality of the probability vectors and a plurality of the compounding ratios, wherein the information processing program further causing the computer to:

cause a dimensionality of the probability vector to match a number of the plurality of second polymers; and

cause a structure of the composite feature vector to be identical to a structure of the probability vector;

calculating a predicted value of a characteristic of the polymer alloy by inputting the composite feature vector into a third trained machine learning model, wherein the predicted value indicates at least one of glass transition temperature and elastic modulus; and

causing a display device to display the predicted value.