[TRUSTED DOWNLOAD] Adaptive Filtering Algorithms and Practical Implementation solution manual.rar
This survey paper aims to provide a researcher interested in transfer learning with an overview of related works, examples of applications that are addressed by transfer learning, and issues and solutions that are relevant to the field of transfer learning. This survey paper provides an overview of current methods being used in the field of transfer learning as it pertains to data mining tasks for classification, regression, and clustering problems; however, it does not focus on transfer learning for reinforcement learning (for more information on reinforcement learning see Taylor ). Information pertaining to the history and taxonomy of transfer learning is not provided in this survey paper, but can be found in the paper by Pan . Since the publication of the transfer learning survey paper by Pan  in 2010, there have been over 700 academic papers written addressing advancements and innovations on the subject of transfer learning. These works broadly cover the areas of new algorithm development, improvements to existing transfer learning algorithms, and algorithm deployment in new application domains. The selected surveyed works in this paper are meant to be diverse and representative of transfer learning solutions in the past 5 years. Most of the surveyed papers provide a generic transfer learning solution; however, some surveyed papers provide solutions that are specific to individual applications. This paper is written with the assumption the reader has a working knowledge of machine learning. For more information on machine learning see Witten . The surveyed works in this paper are intended to present a high-level description of proposed solutions with unique and salient points being highlighted. Experiments from the surveyed papers are described with respect to applied applications, other competing solutions tested, and overall relative results of the experiments. This survey paper provides a section on heterogeneous transfer learning which, to the best of our knowledge, is unique. Additionally, a list of software downloads for various surveyed papers is provided, which is unique to this paper.
There are different strategies and implementations for solving a transfer learning problem. The majority of the homogeneous transfer learning solutions employ one of three general strategies which include trying to correct for the marginal distribution difference in the source, trying to correct for the conditional distribution difference in the source, or trying to correct both the marginal and conditional distribution differences in the source. The majority of the heterogeneous transfer learning solutions are focused on aligning the input spaces of the source and target domains with the assumption that the domain distributions are the same. If the domain distributions are not equal, then further domain adaptation steps are needed. Another important aspect of a transfer learning solution is the form of information transfer (or what is being transferred). The form of information transfer is categorized into four general Transfer Categories . The first Transfer Category is transfer learning through instances. A common method used in this case is for instances from the source domain to be reweighted in an attempt to correct for marginal distribution differences. These reweighted instances are then directly used in the target domain for training (examples in Huang , Jiang ). These reweighting algorithms work best when the conditional distribution is the same in both domains. The second Transfer Category is transfer learning through features. Feature-based transfer learning approaches are categorized in two ways. The first approach transforms the features of the source through reweighting to more closely match the target domain (e.g. Pan ). This is referred to as asymmetric feature transformation and is depicted in Fig. 1b. The second approach discovers underlying meaningful structures between the domains to find a common latent feature space that has predictive qualities while reducing the marginal distribution between the domains (e.g. Blitzer ). This is referred to as symmetric feature transformation and is depicted in Fig. 1a. The third transfer category is to transfer knowledge through shared parameters of source and target domain learner models or by creating multiple source learner models and optimally combining the reweighted learners (ensemble learners) to form an improved target learner (examples in Gao , Bonilla , and Evgeniou ). The last transfer category (and the least used approach) is to transfer knowledge based on some defined relationship between the source and target domains (examples in Mihalkova  and Li ).
The majority of transfer learning solutions surveyed are complex and implemented with non-trivial software. It is a great advantage for a researcher to have access to software implementations of transfer learning solutions so comparisons with competing solutions are facilitated more quickly and fairly. Table 5 provides a list of available software downloads for a number of the solutions surveyed in this paper. Table 6 provides a resource for useful links that point to transfer learning tutorials and other interesting articles on the topic of transfer learning.