Research On Knowledge Based Statistical Invariant Learning Model

Knowledge is an indispensable part of data research,especially in the process of machine learning,knowledge is particularly important,they can let the machine glimpse the correlation,can enable the machine to better use the principle,so as to make the model more effective,so that the experiment saves more time and energy,but also ensure the effectiveness of the experimental results.The recently proposed approach to Learning using statistical invariants provides a new learning paradigm for classification problems that minimizes desired errors by constructing and maintaining corresponding statistical invariants,which differs from the classical paradigm.This method approximates the expected function using the strong convergence mechanism and the weak convergence mechanism from Hilbert space.While LUSI is suitable for classification,it is very important to select valid predicates to introduce invariants for a particular problem;inappropriate predicates do not solve the problem effectively.How to construct the corresponding statistical invariants based on knowledge to improve the generalization performance of the model is a problem worth studying.Therefore,this paper proposes a knowledge-based use statistics invariant learning method to solve the problem of class imbalance and privilege information.The specific research content of this paper is as follows:(1)This paper calculated appropriate predicates and constructed corresponding statistical invariants for the model based on the knowledge of class unbalance problem and combined with LUSI’s method.In addition,the paper proposes an approach to solve category unbalance with invariants,Learning using rebalanced statistical invariants,which adds different "rebalancing" heuristics to the LUSI model.The importance of probability label information(knowledge)in classification is fully considered,and the unbalanced block of V matrix in LUSI is effectively solved by combining the rebalancing statistical invariant of invariant in rebalancing environment.(2)Appropriate use of privileged information can enhance the training effect of the model.Based on the guiding role of privileged information in the training process,this paper integrates it with statistical invariants in the LUSI model,proposes appropriate predicates and constructs corresponding statistical invariants.In order to improve the training effect of the model and the classification performance of the classifier,the privileged information is considered when the model is selecting the admissible set.(3)In order to solve the objective equation of LUSI’s method,Lagrange multipliers are also used to solve LUSI’s method in this paper.Finally,a large number of experiments are conducted on real data sets to verify the effectiveness of appropriate predicates on unbalanced data and data with privileged information.In addition,this paper also analyzes the performance of the proposed r LUSI method and LUSI method in dealing with unbalance problems.Preliminary experimental results show that it is important to use appropriate invariants,and statistical invariants combined with knowledge can effectively improve the generalization performance of LUSI.