In recent years, AI has been constructed that inputs an integrand and outputs a primitive function, making it possible to integrate mathematical formulas that could not be solved on a computer so far.However, with the conventional method, information such as the arithmetic rules and order of mathematical formulas cannot be reflected in AI, and AI optimized for integration has not been constructed.

　The research group noticed that the mathematical processing of integrals learned in high school is similar to transformations such as language translation by AI, which has been rapidly developing in recent years, and the integrated function when an integrated function (integrated function) is input. We have developed an AI that can predict (primitive function).In addition, based on the idea that it is possible to judge the correctness of the integral by differentiating the primitive function issued by AI and determining whether it matches the integrand, various AIs are constructed and learned to obtain the correct answer. I devised a method to adopt what I was able to put out.

　As a result, it was shown that the implemented AI can be integrated with an accuracy of 99.79%, achieving the highest accuracy compared to the integration tools Mathematica and machine learning-based methods developed so far.Furthermore, by investigating the characteristics of the mathematical formulas learned by AI, it becomes clear that there are functions that are good at integrating and functions that are not good at integrating for each constructed AI. Was able to achieve high accuracy.

　Integral is an essential process for simulations in control engineering and systems biology, and this achievement is expected to contribute to more accurate simulations in these fields.

Paper information:[IEEE Access] Symbolic Integration by Integrating Learning Models With Different Strengths and Weaknesses