Multi-label Determination Of Feature Combination Data Which Are Not In Training Set Using Single Neural Network

Purpose: To explore a kind of algorithm combined with deep learning algorithms,so that in the multi-label classification,the combination of data labels that have never appeared in the training set appears in the test set.Using this algorithm can make a more accurate judgment and use as few as possible time.This is an improvement on the traditional deep learning multi-label learning algorithm,and it also provides a new idea for the deep learning interpretability.Method: Intercept the characteristics of the middle layer of the neural network,use linear method to reduce the dimension,calculate the probability distribution of the points after the dimension reduction,and calculate the Euclidean distance between the data point and each single label classification center,so as to determine whether this data belongs to a single label or Different label combinations.Results: Through the middle layer of the network,it is found that there is a certain degree of linear relationship between different label combinations and a single label.Using the algorithm provided in this paper,the network can determine the label combination that does not exist in the training set.This is a multi-label that previously used deep learning.The algorithm cannot.This method requires only one neural network,which greatly saves computing time.Conclusion: The neural network can reflect the distribution of different features,and can draw conclusions based on the features of the middle layer.Although the accuracy of the calculation needs to be improved,for the moment,it has proved to be feasible.This algorithm is somewhat innovative.Different from the original algorithm,it can be inferred that the training set does not contain conditions,indicating that the features of deep learning can be clustered and reflect the characteristics of the original data to a certain extent,which can be explained by deep learning.Contributions to the study of sexual theory.