Domain adaptation on graphs by learning aligned graph bases

Download
index.pdf
2018
Pilancı, Mehmet
Show full item record
173
views
137
downloads
In this thesis, the domain adaptation problem is studied and a method for domain adaptation on graphs is proposed. Given sufficiently many observations of the label function on a source graph, we study the problem of transferring the label information from the source graph to a target graph for estimating the target label function. Our assumption about the relation between the two domains is that the frequency content of the label function, regarded as a graph signal, has similar characteristics over the source and the target graphs. We propose a method to learn a pair of coherent bases on the two graphs, such that the corresponding source and target graph basis vectors have similar spectral content, while “aligning” the two graphs at the same time so that the reconstructed source and target label functions have similar coefficients over the bases. We formulate the basis learning problem as the learning of a linear transformation between the source and target graph Fourier bases so that each source Fourier basis vector is mapped to a new basis vector in the target graph obtained as a linear combination of the target Fourier basis vectors. One synthetic dataset, two image datasets and one book review dataset are used to test the performance of the proposed algorithm. Besides, baseline machine learning methods and recent domain adaptation algorithms are utilized to compare the performance of the proposed algorithm with the methods in the literature. Experiments on several types of data sets suggest that the proposed method compares quite favorably to reference domain adaptation methods. To the best of our knowledge, our treatment is the first to study the domain adaptation problem in a purely graph-based setting with no need for embedding the data in an ambient space. This feature is particularly convenient for many problems of interest concerning learning on graphs or networks.
Graph theory., Fourier analysis., Laplacian matrices.
http://etd.lib.metu.edu.tr/upload/12622105/index.pdf
https://hdl.handle.net/11511/27296
Graduate School of Natural and Applied Sciences, Thesis

Suggestions

DOMAIN ADAPTATION VIA TRANSFERRING SPECTRAL PROPERTIES OF LABEL FUNCTIONS ON GRAPHS
Pilanci, Mehmet; Vural, Elif (2016-07-12)
We propose a domain adaptation algorithm that relies on a graph representation of data samples in the source and target domains. The algorithm combines the information of the known class labels in the source and target domains through the Fourier coefficients of the class label function in the two graphs. The proposed method does not require an ordering or a one-to-one mapping between the samples of the source and target domains; instead, it uses only a small set of matched pairs that serve the purpose of "...
Domain Adaptation on Graphs via Frequency Analysis
Pilancı, Mehmet; Vural, Elif (2019-08-22)
Classical machine learning algorithms assume the training and test data to be sampled from the same distribution, while this assumption may be violated in practice. Domain adaptation methods aim to exploit the information available in a source domain in order to improve the performance of classification in a target domain. In this work, we focus on the problem of domain adaptation in graph settings. We consider a source graph with many labeled nodes and aim to estimate the class labels on a target graph wit...
Refining technique for multilevel graph k-partitioning and its application on domain decomposition non overlapping Schwarz technique for urban acoustic pollution
Faucard, B.; Sorguç, Arzu; F., Magoules; I., Hagiwara (2009-06-01)
It is known that convergence rate of Domain Decomposition Method for finite element analysis depends on the mesh decomposition quality. In this study, a new approach for refinement algorithm to be employed in mesh segmentation is presented Method is based on multilevel quadrisection and octasection graph partitioning. In this method, the connection is guaranteed first by detecting the disconnected parts and then through the algorithm, assigning them to the domain which mostly contain the mesh of interest. H...
Learning Graph Signal Representations with Narrowband Spectral Kernels
Kar, Osman Furkan; Turhan, Gülce; Vural, Elif (2022-01-01)
In this work, we study the problem of learning graph dictionary models from partially observed graph signals. We represent graph signals in terms of atoms generated by narrowband graph kernels. We formulate an optimization problem where the kernel parameters are learnt jointly with the signal representations under a triple regularization scheme: While the first regularization term aims to control the spectrum of the narrowband kernels, the second term encourages the reconstructed graph signals to vary smoot...
Some remarks on vector-valued integration
Kadets, Vladimir; Shumyatskiy, Boris; Shvidkoy, Roman; Zheltukhın, Kostyantyn; Tseytlin, Leonid (2002-02-01)
The paper continues the study of the notion of Riemann–Lebesgue integral, which was introduced before by two of the authors. The result about the convexity of the limit set of integral sums is generalized to the case of weakly-compactly generated spaces. The notion of Riemann–Lebesgue integral is used to introduce new classes of Banach spaces. The properties of these new spaces are studied.
Citation Formats
M. Pilancı, “Domain adaptation on graphs by learning aligned graph bases,” M.S. - Master of Science, Middle East Technical University, 2018.
METU IIS - Integrated Information Service open@metu.edu.tr