Encoding and decoding method and device for a machine learning model

Abstract

The invention discloses an encoding and decoding method and a device of a machine learning model. The encoding and decoding method and the device comprise the following steps: collecting original dataand sampling the original data; The original data is encoded into model code by model generator. A model decoder is used to decode the model-encoded sampling results into valid models. Use raw data to train valid models. The technical proposal of the invention can encode and decode different machine learning models or different types of machine learning models, reduce computational complexity, and quickly generate machine learning models.

Description

technical field [0001] The present invention relates to the field of machine learning, and more specifically, to a coding and decoding method and device for a machine learning model. Background technique [0002] Currently, machine learning is not widely used. First, the modeling process is cumbersome. Algorithm modeling and parameter tuning is a very cumbersome process that requires continuous iterations of data preprocessing, feature engineering, model parameter tuning, and model evaluation to complete the complex modeling process. A data modeling team of 3-6 people Complex modeling often takes months to complete. Second, there is a shortage of artificial intelligence talents. Machine learning involves a lot of engineering and algorithms, and relies on the resources of scientists. However, the current AI talent gap in China exceeds 5 million, with a supply-demand ratio of only 1:10, and artificial intelligence professionals generally have higher salaries, bringing huge ...

AI Technical Summary

Problems solved by technology

[0004] Since the essential work of AutoML is to select, combine, and optimize the methods of each process, there are the following problems: we usually do not know what kind of display expression between the optimized parameters and the expected effect, so the form of the objective function is unknown; There are too many possible combinations, so the search space is huge; there are too many combinations, and each combination needs to do data preprocessing, feature processing, model training and other operations from scratch, so the cost of function calculation is huge

[0005] Aiming at the problem that the computational complexity of building a model in the prior art is too large and difficult to implement, there is currently no effective solution

Images