Information Technology Reference
In-Depth Information
(a) PIE CMU dataset
(b) COIL dataset
(c) USPS dataset
(d) Adult dataset
(e) Text dataset
(f) colt98 dataset
Fig. 3. Accuracy of different datasets
in Table 2. The dimension reduction result project data into 3D by computing
KSIR with small amount and entire labeled instances are shown in Figures 2.
It illustrates that small amount labeled data is enough to estimated a e.d.r.
subspace.
In order to evaluate the overall performance of our semi-supervised KISR
method, we compare it with several state-of-the-art semi-supervised algorithms
including the RSVM based two-teachersone-student semi-supervised learning al-
gorithm (2T1S) in[1] and low density separation (LDS) in [2] and a pure super-
vised benchmark given by running a nonlinear SSVM [6] classifier trained on
Search WWH ::




Custom Search