Interpretable symbolic decomposition of numerical coefficients

A computer system comprising:

• a memory;
• and
• a processor system communicatively coupled to the memory;
• the processor system configured to perform a method comprising:
• receiving, by a machine learning system based on a neural network comprising one or more simulated neurons, a numerical coefficient to decompose into at least one mathematical expression;
• receiving, from a user, an input that identifies at least one relevant numerical constant, wherein the identified at least one relevant numerical constant is determined to be likely included within an accurate decomposition of the numerical coefficient, the at least one relevant numerical constant is assigned a lower complexity cost, wherein the at least one relevant numerical constant corresponds to a technological field of the user, and wherein assigning the lower complexity cost for the at least one relevant numerical constant comprises adjusting at least one numeric weight of the one or more simulated neurons;
• receiving, from the user, a second input that identifies at least one relevant operator, wherein the identified at least one relevant operator is determined to be likely included within an accurate decomposition of the numerical coefficient, and the at least one relevant operator is assigned a lower complexity cost;
• receiving, from the user, a third input comprising a grammatical rule set defining grammar of the at least one mathematical expression, the grammatical rule set corresponding to the technological field of the user;
• decomposing the numerical coefficient into the at least one mathematical expression, wherein decomposing takes into account a complexity cost of the at least one mathematical expression and the grammatical rule set;
• and
• generating an expression tree representation of the decomposing of the numerical coefficient into the at least one mathematical expression and outputting the expression tree representation and the at least one mathematical expression.

A computer program product comprising a computer-readable storage medium having program instructions embodied therewith, the program instructions readable by a processor system to cause the processor system to:

• receive, by a machine learning system based on a neural network comprising one or more simulated neurons, a numerical coefficient to decompose into at least one mathematical expression;
• receive, from a user, an input that identifies at least one relevant numerical constant, wherein the identified at least one relevant numerical constant is determined to be likely included within an accurate decomposition of the numerical coefficient, the at least one relevant numerical constant is assigned a lower complexity cost, wherein the at least one relevant numerical constant corresponds to a technological field of the user, and wherein assigning the lower complexity cost for the at least one relevant numerical constant comprises adjusting at least one numeric weight of the one or more simulated neurons;
• receive, from the user, a second input that identifies at least one relevant operator, wherein the identified at least one relevant operator is determined to be likely included within an accurate decomposition of the numerical coefficient, and the at least one relevant operator is assigned a lower complexity cost;
• receive, from the user, a third input comprising a grammatical rule set defining grammar of the at least one mathematical expression, the grammatical rule set corresponding to the technological field of the user;
• decompose the numerical coefficient into the at least one mathematical expression, wherein decomposing takes into account a complexity cost of the at least one mathematical expression and the grammatical rule set;
• and
• generate an expression tree representation of the decomposing of the numerical coefficient into the at least one mathematical expression and outputting the expression tree representation and the at least one mathematical expression.