JP-7856925-B2 - Estimation program, information processing device, and machine learning method

Inventors

• 竹森 翔
• 梅田 裕平

Assignees

• 富士通株式会社

Dates

Publication Date

20260512

Application Date

20220929

Claims (8)

1. Among the algorithms for calculating the energy corresponding to a molecule through iterative processing, a first algorithm, which is different from a second algorithm that uses quantum circuit data, is executed based on molecular information indicating the molecule to be analyzed, and the first number of iterations of the first algorithm is identified. The number of iterations of the first algorithm is used as the explanatory variable, and the number of iterations of the second algorithm is used as the target variable. The number of iterations of the first algorithm is then input into the first machine learning model, which has been trained with the number of iterations of the first algorithm as the explanatory variable. The system outputs an estimate of the second number of iterations when executing the second algorithm based on the molecular information calculated by the first machine learning model. An estimation program characterized by having a computer perform the processing.
2. The explanatory variable further includes the distances between multiple atoms contained in the molecule, and the input of the first iteration count further includes inputting the first distance indicated by the molecular information. The estimation program according to feature 1.
3. A first feature quantity is identified that indicates the characteristics of the first quantum circuit data used when executing the second algorithm based on the molecular information, The first features are input into a second machine learning model, which is trained with the features of the quantum circuit data as explanatory variables and the unit execution time per iteration included in the iterative process of the second algorithm as the objective variable. The system outputs an estimate of the first unit execution time when the second algorithm is executed based on the molecular information calculated by the second machine learning model. The estimation program according to claim 1, characterized in that it causes the computer to perform further processing.
4. Based on the estimated number of iterations in the second step and the estimated unit execution time in the first step, the execution time when the second algorithm is executed based on the molecular information is estimated. The estimation program according to claim 3, characterized in that it causes the computer to perform further processing.
5. Based on the estimated execution time, a job is scheduled to calculate the energy corresponding to the molecule being analyzed. The estimation program according to claim 4, characterized in that it causes the computer to perform further processing.
6. The first algorithm is an inter-configuration interaction method or a coupled cluster method, and the second algorithm is a variational quantum eigenvalue solver method. The estimation program according to feature 1.
7. A memory unit that stores a first machine learning model, which is trained using an algorithm that iteratively calculates the energy corresponding to a molecule, with the number of iterations of the first algorithm as the explanatory variable and the number of iterations of the second algorithm that uses quantum circuit data as the objective variable. A control unit that executes the first algorithm based on molecular information indicating the molecule to be analyzed, identifies the first number of iterations of the first algorithm, inputs the first number of iterations to the first machine learning model, and outputs an estimated value of the second number of iterations when the second algorithm is executed based on the molecular information, calculated by the first machine learning model. An information processing device characterized by having the following features.
8. Among the algorithms for calculating the energy corresponding to a molecule through iterative processing, a second algorithm that uses quantum circuit data and a first algorithm different from the second algorithm are executed based on molecular information representing the sample molecule, and the first number of iterations of the first algorithm and the second number of iterations of the second algorithm are determined. Using the training data including the number of iterations of the first and second algorithms, a first machine learning model is trained with the number of iterations of the first algorithm as the explanatory variable and the number of iterations of the second algorithm as the dependent variable. A machine learning method characterized by having a computer perform the processing.

Description

The present invention relates to an estimation program, an information processing device, and a machine learning method. Computers sometimes perform molecular simulations to analyze the properties of molecules through numerical calculations. Molecular simulations are used in industrial fields such as materials development and pharmaceutical development. Molecular simulations include quantum chemical calculations that microscopically calculate the energy of molecules based on the electronic state of the molecule and the Schrödinger equation. Algorithms for quantum chemical calculations include those that utilize quantum circuit data, such as the Variational Quantum Eigensolver (VQE). Algorithms using quantum circuit data can also be executed by quantum computers. Furthermore, other algorithms exist for quantum chemical calculations, such as the Configuration Interaction (CI) method and the Coupled Cluster (CC) method. A typical algorithm performs an iterative process, repeatedly calculating the energy of a molecule while changing its electron configuration. The algorithm may continue this iterative process until the energy calculation converges. The algorithm may also search for the electron configuration that minimizes the energy and output this minimum energy as the molecule's ground state energy. Furthermore, in the configuration interaction method, a quantum chemical calculation system has been proposed that dynamically selects some of the molecular orbitals among the multiple molecular orbitals of a molecule and calculates the energy of the molecule based on the electron configuration limited to the selected molecular orbitals. International Publication No. 2022/097298 This is a diagram illustrating the information processing device of the first embodiment.This figure shows an example of the hardware of the information processing device according to the second embodiment.This figure shows a comparison of the accuracy and execution time of different algorithms.This figure shows an example of job scheduling.This figure shows examples of input and output data for time-based and iterative models.This is a block diagram showing examples of functions of an information processing device.This is a flowchart showing an example of a machine learning procedure.This flowchart shows an example of the procedure for estimating execution time.This graph shows an example of the accuracy of execution time estimation. The following description of this embodiment will be made with reference to the drawings. First, the first embodiment will be described. Figure 1 is a diagram illustrating the information processing device of the first embodiment. The information processing device 10 uses a machine learning model to estimate the number of iterations of the quantum chemical calculation algorithm. The information processing device 10 may train a machine learning model or execute an algorithm that estimates the number of iterations. The information processing device 10 may also schedule quantum chemical calculation jobs based on the estimated number of iterations. The information processing device 10 may be a client device or a server device. The information processing device 10 may also be called a computer, estimation device, machine learning device, molecular simulation device, or job scheduler. The information processing device 10 includes a storage unit 11 and a control unit 12. The storage unit 11 may be a volatile semiconductor memory such as RAM (Random Access Memory), or a non-volatile storage such as an HDD (Hard Disk Drive) or flash memory. The control unit 12 is a processor such as a CPU (Central Processing Unit), GPU (Graphics Processing Unit), or DSP (Digital Signal Processor). However, the control unit 12 may also include electronic circuits such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array). The processor executes a program stored in memory such as RAM (which may also be the storage unit 11). The collection of processors may be called a multiprocessor or simply a "processor". The memory unit 11 stores the trained machine learning model 15. The machine learning model 15 may be a linear regression model, a nonlinear regression model, or any other type of machine learning model. The machine learning model 15 is trained with the number of iterations of algorithm 13 as the explanatory variable and the number of iterations of algorithm 14 as the dependent variable. Therefore, the machine learning model 15 estimates the number of iterations of algorithm 14 from the number of iterations of algorithm 13. The explanatory variable may also include other features related to the molecule, such as the distance between multiple atoms contained in the molecule. Algorithms 13 and 14 are algorithms for quantum chemical calculations that calculate the energy corresponding to a molecule through iterative processing. For example, algorithms 13 and 14 repeatedly calculate the energy of a