Browsing by Subject "semi-supervised learning"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Graph Adaptive Knowledge Transfer for Unsupervised Domain Adaptation
    (Springer, 2018) Ding, Zhengming; Li, Sheng; Shao, Ming; Fu, Yun; Electrical and Computer Engineering, School of Engineering and Technology
    Unsupervised domain adaptation has caught appealing attentions as it facilitates the unlabeled target learning by borrowing existing well-established source domain knowledge. Recent practice on domain adaptation manages to extract effective features by incorporating the pseudo labels for the target domain to better solve cross-domain distribution divergences. However, existing approaches separate target label optimization and domain-invariant feature learning as different steps. To address that issue, we develop a novel Graph Adaptive Knowledge Transfer (GAKT) model to jointly optimize target labels and domain-free features in a unified framework. Specifically, semi-supervised knowledge adaptation and label propagation on target data are coupled to benefit each other, and hence the marginal and conditional disparities across different domains will be better alleviated. Experimental evaluation on two cross-domain visual datasets demonstrates the effectiveness of our designed approach on facilitating the unlabeled target task learning, compared to the state-of-the-art domain adaptation approaches.
  • Thumbnail Image
    Item
    Partially-observed models for classifying minerals on Mars
    (IEEE, 2013) Dundar, Murat; Li, Lin; Rajwa, Bartek; Earth Sciences, School of Science
    The identification of phyllosilicates by NASA's CRISM (Compact Reconnaissance Imaging Spectrometer for Mars) strongly suggests the presence of water-related geological processes. A variety of water-bearing phyllosilicate minerals have already been identified by several research groups utilizing spectral enrichment techniques and matching phyllosilicate-rich regions on the Martian surface to known spectra of minerals found on earth. However, fully automated analysis of the CRISM data remains a challenge for two main reasons. First, there is significant variability in the spectral signature of the same mineral obtained from different regions on the Martian surface. Second, the list of mineral confirmed to date constituting the set of training classes is not exhaustive. Thus, when classifying new regions, using a classifier trained with selected minerals and chemicals, one must consider the potential presence of unknown materials not represented in the training library. We made an initial attempt to study these problems in the context of our recent work on partially-observed classification models and present results that show the utility of such models in identifying spectra of unknown minerals while simultaneously recognizing spectra of known minerals.