Unsupervised Heterogeneous Transfer Learning Based On Partial Co-occurrence Data

Heterogeneous transfer clustering can translate knowledge from some related heterogeneous source domains to the target domain without any supervision.Existing works usually use a large amount of complete co-occurrence data in tasks.Owning to the heterogeneity of the data,projection functions are used to map heterogeneous data to a common latent feature subspace.However,in many real applications,it is not practical to collect abundant co-occurrence data,while the available co-occurrence data are always incomplete.Another commonly encountered problem is that the complex structure of real heterogeneous data may result in substantial degeneration in clustering performance.To address these issues,unsupervised heterogeneous transfer learning method is proposed from two aspects.On one hand,the deep structure of the data is mined to learn the features of subspace with high efficiency and low redundancy.On the other hand,a generalized and effective unsupervised heterogeneous transfer learning framework based on autoencoder is proposed.The main research results and innovative work of these two aspects are as follows:1.Heterogeneous transfer clustering method which aims to make full use of partial co-occurrence data are proposed in this paper.It is superior to the existing methods in three facets.First,fully uses the partial co-occurrence data in both source and target domains to learn a latent space,maximally extracting useful knowledge for clustering from limited information.Second,it incorporates multi-layer hidden representations,accurately preserving the complex hierarchical structure of data.Third,it enforces approximately orthogonal constraint in representations,effectively characterizing the latent subspace with minimal redundancy.An efficient algorithm has been derived and implemented to realize the proposed HTCPC.A series of experiments on the real datasets have illustrated the advantage of the proposed approach compared with state-of-the-art methods.2.Autoencoder is applied to unsupervised heterogeneous transfer learning method based on partial co-occurrence data.This paper explores the effectiveness of unsupervised heterogeneous transfer learning framework based on multiple autoencoders and compares their thransfer performance with multiple constraints.A series of experiments on real data sets have proved the validity and rationality of the framework,and indicates the performance difference between the existing unsupervised heterogeneous transfer learning method and the framework.In this paper,the two kinds of proposed unsupervised heterogeneous transfer method,break through the limitation that existing method can only use complete co-occurrence data,and by making full use of the partial co-occurrence data,to explore the deep structure of the complex data,making the leap for unsupervised heterogeneous transfer from multi-layer matrix decomposition which get local optimal solution to deep neural network which get global optimal solution.Moreover,a new angle is provided into unsupervised heterogeneous transfer learning.The two unsupervised heterogeneous transfer learning methods proposed in this paper break through the limitation that existing methods which only use complete co-occurrence data,and realize unsupervised heterogeneous transfer learning from exploring the deep structure of data and improving the stability of the model by making full use of partial co-occurrence data.