Method for Automatic Generation of AI Model Based on Computational Graph Evolution

Abstract

Provided is an AI model automatic generation method based on computational graph evolution. The method mainly comprises the following steps: pre-setting data; utilizing a genetic algorithm operator to generate a first-generation computational graph model, and computing the performance of the model according to a computational graph structure thereof; removing an invalid model and a repeated model, and taking the remaining models as candidate models and reserving same as seeds for the next generation; selecting a number of optimal models; the candidate models generating a new computational graph model by using the genetic algorithm operator; determining whether the new computational graph model generated in the last step has been generated; storing the new model as a computational graph model for a new generation, and determining whether same satisfies the pre-set data and an evolution ending condition; and summarizing evolution computational results, and selecting an optimal model. In the present invention, machine learning and deep learning can be carried out simultaneously; the repeated computation of the same model is prevented, and the model design efficiency is improved; the local optimum is jumped out of; the decline in the ability to search for a network is prevented; and evaluation can be directly carried out without training by means of actual data.

Description

Technical field

[0001] The invention relates to the related technical field of AI models (AI models, namely artificial intelligence models), in particular to a method for automatically generating AI models based on computational graph evolution.Background technique

[0002] The automatic generation of AI models is a frontier research field. Automatic model generation can generate simpler and more efficient neural networks based on the distribution of data. The search space automatically generated by the AI ​​model is fn×2n(n-1) / 2 , Where f is the number of different neuron operators and n is the maximum depth of the neural network. It can be seen that during the generation process, with the increase of the supported neural network operators and the deepening of the network, the complexity of the problem may become a problem approaching an infinite search space, resulting in failure to solve.

[0003] At present, the main search methods include reinforcement learning (ie reinforcement learning...

Images