Spectral clustering and transductive learning with multiple views

Abstract

We consider spectral clustering and transductive inference for data with multiple views. A typical example is the web, which can be described by either the hyperlinks between web pages or the words occurring in web pages. When each view is represented as a graph, one may convexly combine the weight matrices or the discrete Laplacians for each graph, and then proceed with existing clustering or classification techniques. Such a solution might sound natural, but its underlying principle is not clear. Unlike this kind of methodology, we develop multiview spectral clustering via generalizing the normalized cut from a single view to multiple views. We further build multiview transductive inference on the basis of multiview spectral clustering. Our framework leads to a mixture of Markov chains defined on every graph. The experimental evaluation on real-world web classification demonstrates promising results that validate our method.

Transductive Multi-view Embedding for Zero-Shot Recognition and Annotation

“…Classification on multiple graphs (C-MG) is well-studied in semi-supervised learning. Most solutions are based on the seminal work of Zhou et al [38] which generalises spectral clustering from a single graph to multiple graphs by defining a mixture of random walks on multiple graphs. However crucially, the influence/trustworthiness of each graph is given by a weight that has to be pre-defined and its value has a great effect on the performance of C-MG [38].…”

“…Most solutions are based on the seminal work of Zhou et al [38] which generalises spectral clustering from a single graph to multiple graphs by defining a mixture of random walks on multiple graphs. However crucially, the influence/trustworthiness of each graph is given by a weight that has to be pre-defined and its value has a great effect on the performance of C-MG [38]. In this work, we extend the C-MG algorithm in [38] by introducing a Bayesian prior weight for each graph, which can be measured automatically from data.…”

“…However crucially, the influence/trustworthiness of each graph is given by a weight that has to be pre-defined and its value has a great effect on the performance of C-MG [38]. In this work, we extend the C-MG algorithm in [38] by introducing a Bayesian prior weight for each graph, which can be measured automatically from data. Our experiments show that our TMV-BLP algorithm is superior to [38] and [27].…”

“…In this work, we extend the C-MG algorithm in [38] by introducing a Bayesian prior weight for each graph, which can be measured automatically from data. Our experiments show that our TMV-BLP algorithm is superior to [38] and [27]. Our contributions are as follows: (1) To our knowledge, this is the first attempt to investigate and provide a solution to the projection domain shift problem in zero-shot learning.…”

“…To propagate label information from labelled nodes to other target instances, a classic strategy is random walks [38]. We define a random walk process within and across graphs.…”