Model self-adaption method and system based on intelligent computing framework

Abstract

The embodiment of the invention provides a model self-adaption method and system based on an intelligent computing framework. The method comprises the following steps: based on an automatic machine learning technology, realizing automatic feature selection, hyper-parameter optimization and neural network architecture search during model construction; based on an expanded open type neural network exchange technology, realizing model cross-framework migration after model construction. According to the embodiment of the invention, important steps related to features, models, optimization and evaluation are automatically learned through a self-adaptive method during model construction; therefore, the manual intervention degree in the machine learning process is reduced, the machine learning method can be more simply and efficiently used by a user, the algorithm library of each mainstream framework is abstracted, the models are stored in a unified format by means of a neural network exchange technology, cross-framework migration of the models is realized, the reusability of the models is improved, and the development cost is reduced.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a model adaptive method and system based on an intelligent computing framework. Background technique [0002] In recent years, the theory and technology of artificial intelligence have been developed by leaps and bounds, and intelligent algorithms emerge in endlessly. Traditional machine learning algorithms include decision trees, random forests, artificial neural networks, Bayesian learning, etc., based on a large amount of data-driven artificial intelligence algorithms, such as deep learning, reinforcement learning, transfer learning, meta-learning, etc., have been obtained in various fields a wide range of applications. However, the application of machine learning algorithms, especially deep neural networks, often requires a lot of manual intervention, which is mainly manifested in: feature extraction, model selection, parameter adjustment and other aspects. ...

AI Technical Summary

Problems solved by technology

However, the development of ONNX technology is still immature. The current ONNX technology only supports model reasoning, and the imported models need to be trained under the original computing framework, which makes it difficult to train and adjust ONNX model files according to actual application requirements.

Moreover, even if the model can be converted between frameworks using the model framework migration technology, the effect of parameter tuning is still difficult to achieve seamless migration

Images