Recognition And Intelligent Modeling Methods Based On Multi-view Learning And Trasnfer Learning

Artificial intelligence technology has developed for nearly 70 years, during which various intelligent methods have been proposed to solve a variety of problems. Fuzzy recognition technology and fuzzy intelligent modeling technology have been extensively concerned and used in healthcare, management, economy and other fields. However, as people’s living condition and technology improves, more and more new application scenarios are discovered. Among the new application scenarios, multi-view application scenario and transfer application scenario have a broad impact on people’s life and production. This study will mainly focus on the two emerging application scenarios. Researchers find that performance of some classic fuzzy recognition technologies and fuzzy intelligent modeling technologies become no longer reliable in the above two new application scenarios, which usually face the following several challenges: 1) In multi-view application scenarios, classic fuzzy recognition technology and fuzzy intelligent modeling technology are unable to learn collaboratively from multiple views since they are both proposed in learning scenario of single view. If these classic methods are insistently applied to multi-view learning, they can only be learned in each view independently, which results in unsatisfied outcome; 2) In transfer application scenarios, due to the confidentiality of production or a new industry without data accumulated in the past, there is little data available for classic fuzzy recognition technology and fuzzy intelligent modeling technology to handle the industry with data processing and learning modeling, which leads to failure of the classic methods.In order to solve the problems that classic fuzzy recognition technology and fuzzy intelligent modeling technology face in these emerging application scenarios, this study will be divided into two parts to make appropriate improvements for fuzzy recognition technology and fuzzy intelligent modeling technology in multi-view scenario and transfer scenario respectively, viewing to improve performance of the model. Details are as follows:(1) The first part is from the section 2 to section 4, which mainly discusses the application of fuzzy recognition technology and fuzzy intelligent modeling technology in multi-view field. In the beginning, according to problems of an existing multi-view model based on fuzzy clustering algorithm called Co-FCM in section 2, a novel multi-view collaborative learning method is reconstructed based on Havrda-Charvat entropy theory, which is called approximation-criteria of cluster partition. Furthermore, an adaptive weighting strategy for each view based on the theory of shannon entropy is proposed specific to differences between each view in multi-view scenario. Finally, the collaborative partition multi-view fuzzy clustering algorithm using entropy weighting(EW-Co P-MVFCM) is proposed. Secondly, based on awareness of multi-view fuzzy clustering in section 2 and GIFP-FCM algorithm on account of previous work on fuzzy clustering in section 3, a basic algorithm named Co-FCM is raised which is able to learn collaboratively from multiple views. Co-FCM algorithm introduces four different collaborative metric functions to expand the application scope of the method. Moreover, in consideration of the otherness of each view, weighted mechanism of view is added to the basic Co-FCM algorithm, which finally comes up with WV-Co-FCM algorithm. Compare to Co-FCM algorithm, WV-Co-FCM algorithm not only wins the recognition capability of best view and also has better clustering performance. Eventually, in section 4, application of fuzzy intelligent modeling technology in multi-view field is discussed. Specifically, the TSK-FC algorithm for fuzzy classification model of single-view based on classification mechanism of large interval is presented at first. As a base model, the proposed TSK-FC is developed to a TSK model of fuzzy classification named Two V-TSK-FC with double-view synergetic learning capacity, by integrating synergetic learning mechanism of multiply views. By means of synergetic learning mechanism, the model can use the independent information of each view and associated information between each view in further to enhance performance of algorithm during the modeling process.(2) The second part is from section 5 to section 7, which mainly discusses the application of fuzzy recognition technology and fuzzy intelligent modeling technology in transfer learning field. First of all, in section 5, according to the serious influence on final clustering result caused by the insufficient dataset(poor data) and noise, generalized clustering algorithm with improved fuzzy partitions(GIFP-FCM) is mentioned still based on this section. The GIFP-FCM algorithm can obtain the transfer ability by integrating with transfer learning mechanism with clustering features, and eventually comes up with transfer GIFP-FCM algorithm called T-GIFP-FCM. Secondly, for existing problems of traditional fuzzy system modeling in transfer scenario, the fuzzy system with transfer learning capacity is discussed focusing on the widespread used TSK type fuzzy system, which is called TSK-transfer learning fuzzy system. The TSK-transfer learning fuzzy system will not only make full use of the data information in current scenario, but also effectively learn from the knowledge accumulated in the related historical scenario to assist the learning of modeling process in the current scenario, thus to improve generalization performance of the model. Finally, section 7 further puts forward the corresponding improvement program for a series of problems, existing in transfer learning of premise parameters and consequent parameters by TSK-transfer learning fuzzy system raised in section 6. Specifically, combined with the fuzzy transfer clustering theory as well as an improved transfer learning mechanism of consequent presented in section 5, the TSK transfer learning fuzzy system based on the improved knowledge-transfer is proposed. The method can efficiently unite transfer clustering and modeling of transfer fuzzy system, which leads to more intelligent modeling process and more excellent learning ability of fuzzy system. Meanwhile, the proposal of the method provides a new research idea for development of transfer learning in intelligent modeling field.