Calculation system for optimizing hyperparameters based on evolutionary planning algorithms

Abstract

This disclosure provides a computational system comprising a terminal device and a processor. The processor, located in the terminal device, includes a random generation module, an adaptation evaluation module, an individual mutation module, and a population competition module. The random generation module forms an initial population; the adaptation evaluation module, coupled to the random generation module, calculates a fitness value from the initial population; the individual mutation module, coupled to the random generation and adaptation evaluation modules, processes the fitness value to generate a mutant population; and the population competition module, coupled to the random generation, adaptation evaluation, and individual mutation modules, continuously iterates between the initial and mutant populations to generate new surviving populations. This allows for precise adjustment of hyperparameters.

Claims

1. A computational system for optimizing hyperparameters based on an evolutionary programming algorithm, comprising: a terminal device; and a processor disposed on the terminal device, the processor comprising: a random generation module that randomly generates a plurality of individuals, the individuals having a random matrix, the random generation module processing the random matrix according to a hyperparameter evolution equation to form an initial population; an adaptation evaluation module coupled to the random generation module, the adaptation evaluation module calculating the initial population according to an evaluation equation, and the adaptation evaluation module obtaining a fitness value; and an individual mutation module coupled to the random generation module and the adaptation evaluation module, the individual mutation module processing the fitness value of the initial population according to an individual mutation equation, and the individual mutation module generating a mutant population; A population competition module is coupled to the random generation module, the adaptation evaluation module, and the individual mutation module. The population competition module has a termination condition. The population competition module selects a surviving population from the initial population and the mutation population. The population competition module continuously iterates among the surviving population, the initial population, and the mutation population to create new surviving populations until the termination condition of the population competition module is met. At this point, the population competition module stops selecting new surviving populations and outputs the last group of surviving populations to the terminal device.

2. The computational system for optimizing hyperparameters based on evolutionary programming algorithm as described in claim 1, wherein, The population competition module has an early termination condition. The time taken by the population competition module to select the surviving population from the initial population and the mutant population is an iteration time. When the iteration time multiplied by the number of iterations of the surviving population, the initial population and the mutant population exceeds the early termination condition, the population competition module stops selecting a new surviving population in advance.

3. The computational system for optimizing hyperparameters based on the evolutionary programming algorithm as described in claim 2, wherein, The mutant population is defined as follows: when an individual in a certain group survives, the population competition module will form a new surviving population from the surviving individuals; when an individual in a certain group survives, the population competition module will form a new surviving population from the surviving individuals.

4. The computational system for optimizing hyperparameters based on the evolutionary programming algorithm as described in claim 3, wherein, The termination condition is a preset number of iterations. When the number of iterations of the surviving population, the initial population, and the mutated population reaches the preset number, the population competition module stops selecting a new surviving population.

5. The computational system for optimizing hyperparameters based on the evolutionary programming algorithm as described in claim 1, wherein, The terminal device stores electrocardiogram (ECG) image data, and the random generation module processes the ECG image data.

Images