Malware is defined as any type of computer software harmful to computers or networks,which has been posing a serious threat to the global security.Whats more,the amount of malware is increasing rapidly in recent years.Therefore,detecting malicious code is of great significance and draw attention of experts worldwide in the field of information security.Due to huge amount of malware,it is impractical to detect all malicious code manually,which lead to the application of Supervised machine learning models.To some extent,these Supervised machine learning models such as SVM,J48 alleviate the workload of human.However,these algorithms still require a great number of labeled samples including malicious code and benign software to train models beforehand.In order to further reduce the number of samples requiring to be labeled remarkably,some experts propose new methods to detect unknown malware with the help of semi-supervised learning approaches such as self-train and collaborative train.Although these methods work well in many situation,some issues limit the extensive applying.The effect of collaborative train decreases evidently when the feature is single-view.In this paper,we propose to release the limit of single-view with the help of ICA(independent component analysis),and bring out a new method of malware detection that adopts ICA and collaborative train for the first time.With the help of our method,far fewer labeled samples is needed than when supervised learning is used,while the accuracy rate keeps at a high level.