A survey on heterogeneous transfer learning

Journal of Big Data

Table 2 Surveyed HTL methods which require limited target labels and accept unlabeled target data

Methods	Characteristics	Sections
DAMA [27]	Symmetric, manifold alignment with labels, multi-source	“DAMA”
CDLS [46]	Symmetric and landmark weights	“CDLS”
IDL for HDA [49]	Online tasks, symmetric eigenanalysis-based	“IDL for HDA”
MOMAP [51]	Asymmetric, mapping by rotation and translation, multi-class, multi-source	“MOMAP”
HeMap [26]	Symmetric, spectral mapping Bayesian method, cluster-based sampling	“HeMap”
Proactive HTL [54]	Symmetric, label embeddings, proactive learning	“Proactive HTL”
SHFA [47]	Symmetric transformation w/augmentation for semi-supervised	“SHFA”
CT-Learn [57]	Requires co-occurrence data, joint transition probability graph, Markov random walk, multi-source	“CT-Learn”
SSKMDA [60]	Instance-based asymmetric, kernel matching	“SSKMDA”
SCP-ECOC [62]	Symmetric, multi-class, ECOC scheme	“SCP-ECOC”
MMDT [48]	Asymmetric, image, max-margin, multi-class	“MMDT”
SSMVCCAE [64]	Symmetric, multi-view ensemble, CCA analysis, SRKDA	“SMVCCAE, SSMVCCAE”
TNT [65]	Neural network-based mapping and classification	“TNT”
HDANA [67]	Symmetric, deep learning, autoencoder mapping	“HDANA”