Safety semi-supervised extreme learning machine classification method based on collaborative representation

Abstract

The invention relates to a safety semi-supervised extreme learning machine classification method based on collaborative representation, which comprises the following steps: firstly, defining a risk item of an unmarked sample, and judging the risk of the sample by utilizing the consistency of output prediction tags before and after the reconstruction of the unmarked sample and a reconstruction error. According to the invention, a collaborative representation method is adopted to analyze the risk of a label-free sample in semi-supervised learning. Existence of sample risks is verified through experiments; adding the risk item into a loss function of the semi-supervised extreme learning machine. Therefore, the model has a sample security risk strategy. Finally, testing is carried out under the electroencephalogram signal data set, the result shows that when the performance of a traditional semi-supervised extreme learning machine is lower than that of a supervised extreme learning machine, the performance of the safe semi-supervised extreme learning method is still superior to that of a supervised learning method, and it is also proved that the method has certain safety of unmarked samples.

Description

technical field [0001] The invention belongs to the field of pattern recognition, and is a method for analyzing the risk of unlabeled samples by using a cooperative representation method, and adding a risk mechanism to a semi-supervised ultra-limit learning machine to classify electroencephalogram signals. Background technique [0002] Electroencephalogram (EEG) analysis is an effective method for studying brain science, and it provides the possibility for the realization of brain-computer interface (BCI) systems, and the classification of EEG signals based on motor imagery is a typical paradigm. Over the past few years, BCI has provided assistance to patients with disturbance of consciousness and stroke. EEG signals have properties such as non-stationarity and individual differences, which make it difficult for traditional methods to classify accurately, and there are serious time-consuming problems. In recent years, in order to find a high-performance and low-time-consumi...

AI Technical Summary

Problems solved by technology

[0006] The above methods have made good use of various mechanisms for the research of security semi-supervised algorithms, but many methods are based on the risk calculation of the entire training model, and there are few strategies designed for the information of the sample itself, and the application In the classification of EEG signals, this paper proposes a risk strategy based on sample reconstruction and applies it to semi-supervised learning machines, and uses SS-ELM with fast training speed as the semi-supervised classifier in this paper for EEG signal analysis

Images