A survey of transfer learning

Weiss, Karl R.; Khoshgoftaar, Taghi M.; Wang, Dingding

doi:10.1186/s40537-016-0043-6

Cited by 3,986 publications

(2,074 citation statements)

References 94 publications

Supporting

Mentioning

2,025

Contrasting

Unclassified

Order By: Relevance

“…Several machine learning problems fall in this category, probably the most popular are concept drift [Gama et al 2014], domain adaptation [Daume III and Marcu 2006] and transfer learning [Pan and Yang 2010;Weiss et al 2016]. All these problems share that the distribution used for training our model is different than the data the model will face later.…”

Section: Changes In Data Distributionmentioning

confidence: 99%

A Review on Quantification Learning

et al. 2017

View full text Add to dashboard Cite

The task of quantification consists in providing an aggregate estimation (e.g. the class distribution in a classification problem) for unseen test sets, applying a model that is trained using a training set with a different data distribution. Several real-world applications demand this kind of methods that do not require predictions for individual examples and just focus on obtaining accurate estimates at an aggregate level. During the past few years, several quantification methods have been proposed from different perspectives and with different goals. This paper presents a unified review of the main approaches with the aim of serving as an introductory tutorial for newcomers in the field.

show abstract

Section: Changes In Data Distributionmentioning

confidence: 99%

A Review on Quantification Learning

et al. 2017

View full text Add to dashboard Cite

show abstract

“…, OSELM requires a number of instances not less than the hidden neuron number before online training in 1 Θ to form an initialization set, which would greatly affect OSELM's learning capability. Since training instances in 1 Θ are not enough to form a proper initialization set, we turn to Ω for exploiting extra information.…”

Section: The Proposed Otelmmentioning

confidence: 99%

“…If we can pick out these instances from the initialization set, the performance would be improved. Based on this idea, we propose OTELM which consists of three steps: Firstly, with the guidance of 1 N labeled instances in 1 Θ , a SSELM is performed on instances from 1 Θ Ω, which assume instances in Ω are unlabeled and would later be re-labeled after the model is trained. Secondly, pick out instances in Ω whose labels predicted by SSELM are not equal to their real labels.…”

Section: The Proposed Otelmmentioning

confidence: 99%

“…Transfer learning [1] is quite attractive when there are not sufficient instances in the target domain for supervised learning while plenty of instances from a related but different domain (source domain) are available and can be used for improving the final performance. Previous works have investigated to build connections and transfer knowledge from the source to the target domain.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Online transfer learning with extreme learning machine

Yin¹,

Yang²

2017

AIP Conference Proceedings

View full text Add to dashboard Cite

“…Transfer Learning has been employed in a number of domains containing multiple sources to allow data inference to unseen sources. For a more in-depth discussion of the wider field, [6] provides a recent, thorough survey. More specifically, BCI literature typically reports domain adaptation approaches [5], the most popular of which being Common Spatial Patterns [7].…”

Section: Related Work On Transfer Learning In Bcimentioning

confidence: 99%

Evolving Training Sets for Improved Transfer Learning in Brain Computer Interfaces

Adair

Brownlee

Daolio

et al. 2017

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract.A new proof-of-concept method for optimising the performance of Brain Computer Interfaces (BCI) while minimising the quantity of required training data is introduced. This is achieved by using an evolutionary approach to rearrange the distribution of training instances, prior to the construction of an Ensemble Learning Generic Information (ELGI) model. The training data from a population was optimised to emphasise generality of the models derived from it, prior to a re-combination with participant-specific data via the ELGI approach, and training of classifiers. Evidence is given to support the adoption of this approach in the more difficult BCI conditions: smaller training sets, and those suffering from temporal drift. This paper serves as a case study to lay the groundwork for further exploration of this approach.

show abstract

A survey of transfer learning

Cited by 3,986 publications

References 94 publications

A Review on Quantification Learning

A Review on Quantification Learning

Online transfer learning with extreme learning machine

Evolving Training Sets for Improved Transfer Learning in Brain Computer Interfaces

Contact Info

Product

Resources

About