Description
TitleUnsupervised visual domain adaptation: a probabilistic approach
Date Created2020
Other Date2020-01 (degree)
Extent1 online resource (xvii, 109 pages) : illustrations
DescriptionArtificial intelligent and machine learning technologies have already achieved significant success in various applications (computer vision, natural language processing, speech recognition, etc.). Such methods work well only under a common assumption that training and test data are drawn from the same distribution. However, the curse of domain mismatch arises when the test data and the training data come from different distributions. In such distribution changes, most statistical models need to be rebuilt, using newly collected training data. In many real world applications, it is expensive or even impossible to collect the required training data and rebuild the models. One of the ultimate goals of the open ended learning systems is to take advantage of previous experience/ knowledge in dealing with similar future problems. Two levels of learning can be identified in such scenarios. One draws on the data by capturing the pattern and regularities which enables reliable predictions on new samples. The other starts from an acquired source of knowledge and focuses on how to generalise it to a new target concept; this is also known as transfer learning which is going to be the main focus of this thesis.
This thesis will focus on a family of transfer learning methods applied to the task of visual object recognition, specifically image classification. The visual recognition problem is central to computer vision research: many desired applications, from robotics to information retrieval, demand the ability to correctly identify categories, places, and objects. Transfer learning is a general term, and specific settings have been given specific names: when the learner has access to only unlabeled data from the target domain (where the model should perform) and labeled data from a different domain (the source), the problem is called unsupervised domain adaptation (DA).
The thesis focuses on four methods for this setting. The first one proposes a probabilistic latent variable model by learning projections from each domain to a latent (shared) space jointly with the classifier in the latent space, which simultaneously minimizes the domain disparity while maximizing the classifier’s discriminative power. Furthermore, the non-parametric nature of our adaptation model makes it possible to infer the latent space dimension automatically from data.
The second method is based on the Gaussian Process (GP): The GP allows us to induce a hypothesis space of classifiers from the posterior distribution of the latent random functions, turning the learning into a large-margin posterior separation problem.
The Third method is based on GANs: We introduce an adversarial discriminative discrepancy measure which takes advantage of auxiliary information available in the source and the target domains to better align the source and target distributions. Specifically, we leverage the cohesive clustering structure within individual data manifolds, associated with different tasks, to improve the alignment.
The last one addresses domain adaptation for multiple target domains. We propose an information theoretic approach for domain adaptation in the novel context of multiple target domains with unlabeled instances and one source domain with labeled instances. Our model aims to find a shared latent space common to all domains, while simultaneously accounting for the remaining private, domain-specific factors. Disentanglement of shared and private information is accomplished using a unified information-theoretic approach, which also serves to establish a stronger link between the latent representations and the observed data.
We conduct experiments on a wide range of image classification tasks. We test our proposed methods and show that, in all cases, leveraging knowledge from a related domain can improve performance when there are no labels available for direct training on the new target data.
NotePh.D.
NoteIncludes bibliographical references
Genretheses, ETD doctoral
LanguageEnglish
CollectionSchool of Graduate Studies Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.