Equipment authentication method and system for rejecting inference based on shallow self-learning algorithm, and electronic equipment

Abstract

The invention discloses an equipment authentication method and system for rejecting inference based on a shallow self-learning algorithm, and electronic equipment. The method comprises the steps of adopting an unsupervised abnormal data recognition algorithm, eliminating extreme data and unlabeled data with too little information gain, and guaranteeing the stability of self-learning; marking a label-free sample through an iteration threshold quantile; and retraining and sorting a model by combining the label-free sample with the determined label and a passing sample with a label so as to achieve the effect of correcting the grading distribution of the model. According to the invention, human subjective assume is avoided to the greatest extent by using a machine learning algorithm, and a new evaluation standard is provided for evaluating the effectiveness of rejection inference, so that the robustness and the accuracy of equipment authentication are greatly improved.

Description

technical field [0001] The present invention relates to the technical field of equipment authentication on an Internet platform, and more specifically relates to a device authentication method and system based on shallow self-learning algorithm-based rejection inference, an electronic device, and a computer storage medium. Background technique [0002] For mobile applications based on Internet platforms, if users want to handle specific network services, they often need to authenticate the connected devices, and only authenticated devices are eligible to apply for related services. In order to avoid risks, the platform often needs to pre-evaluate new devices, and only devices that pass the evaluation are allowed to apply for subsequent business applications. For example, for travel services based on Internet platforms (taxi-hailing, shared bicycles, navigation), food delivery services (calling express delivery, ordering food delivery), financial services (financial managemen...

AI Technical Summary

Problems solved by technology

However, in one type of machine iteration, due to continuous strategy iteration, only the data that passes certain conditions can obtain the label value, so the scoring model can only be established in the acceptance samples that pass the conditions, and it is different from the actual application in all samples (rejection + Pass) distribution is different, thus causing the problem of sample bias

[0004] The method to solve this kind of sample deviation problem is usually called the rejection inference method. The traditional rejection inference method includes, for example, the hard truncation method of inference with the help of threshold and ratio, and the outsourcing method, etc., but these methods have more or less shortcomings. , such as too mechanical, or too biased towards subjective conjecture, there is an urgent need to research and develop a more accurate rejection inference method

Images