Heterogeneous Transfer Learning for Image Clustering via the SocialWeb

By Qiang Yang, Yuqiang Chen, Gui-Rong Xue, Wenyuan Dai, Yong Yu. 2009

180 citations recorded by the source corpus.

In this paper, we present a new learning scenario, heterogeneous transfer learning, which improves learning performance when the data can be in different feature spaces and where no correspondence between data instances in these spaces is provided. In the past, we have classified Chinese text documents using English training data under the heterogeneous transfer learning framework. In this paper, we present image clustering as an example to illustrate how unsupervised learning can be improved by transferring knowledge from auxiliary heterogeneous data obtained from the social Web. Image clustering is useful for image sense disambiguation in query-based image search, but its quality is often low due to imagedata sparsity problem. We extend PLSA to help transfer the knowledge from social Web data, which have mixed feature representations. Experiments on image-object clustering and scene clustering tasks show that our approach in heterogeneous transfer learning based on the auxiliary data is indeed effective and promising.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Domain Adaptation for Visual Applications: A Comprehensive Survey (2017, methodology): “Fig. 27 Illustrating through an example the difference between TL to ML in the case of homogeneous data and between multi-view and HTL/HDA when working with heterogeneous data. Image courtesy Q. Yang [262]. between the training data from two image categories, and these metrics are used to build a cr”
Learning with Augmented Features for Heterogeneous Domain Adaptation (2012, methodology): “The pioneer works (Dai et al., 2009; Prettenhofer & Stein, 2010; Wei & Pal, 2010; Yang et al., 2009; Zhu et al., 2011) are limited to some specific HDA tasks, because they required additional information to transfer the source knowledge to the target domain.”
Learning With Augmented Features for Supervised and Semi-Supervised Heterogeneous Domain Adaptation (2014, background): “Thus, they cannot deal with the problem where the dimensions of data from the source and target domains are different, which is known as heterogeneous domain adaptation (HDA) [8], [9].”
A Survey of Multi-View Representation Learning (2016, unknown): “Similarly, through multi-modal embedding the complementary knowledge underlying text and image data can be unified to improve the prediction performance by using the complementarity principle.”
Towards semantic knowledge propagation from text corpus to web images (2011, background): “In contrast to previous work, [7][18][17], the translator proposed in this paper can directly establish the semantic correspondence between text and images even if they are new instances of the image data with unknown correspondence to the text documents, or if the co-occurrence data is independent ”
Heterogeneous Transfer Learning for Image Classification (2011, background): “This problem has been referred to as a Heterogeneous Transfer Learning problem (Yang et al. 2009) 1 .”
Transfer Learning on Heterogenous Feature Spaces via Spectral Transformation (2010, background): “[6]) proposed to apply image tags to align image data and text data; but the algorithms rely on image tags to build up the alignment and are especially designed for image and text data, which is not for general case.”
TrGraph: Cross-Network Transfer Learning via Common Signature Subgraphs (2015, background): “TrAdaBoost [24] is one typical example of such methods, which assigns larger weights to the instances from the source domain that are more similar to the target instances.”
Neural-Symbolic Learning and Reasoning: Contributions and Challenges (2015, background): “Some work on heterogeneous transfer learning has been directed at these questions (Yang et al, 2009).”
Instilling Social to Physical: Co-Regularized Heterogeneous Transfer Learning (2016, background): “(Yang et al. 2009) first proposed heterogeneous transfer learning which transfers knowledge across domains in different feature spaces.”
Lifelong Machine Learning Systems: Beyond Learning Algorithms (2013, background): “provides a general reference framework for an LML system that uses universal knowledge and knowledge of the task domain as a source of inductive bias (Silver and Mercer 2002; Yang et al. 2009).”