Special Issue on Small Data AnalyticsInternational Journal of Machine Learning and CyberneticsOpen for submissions until May 31 2021
The IssueBig data is an important characteristic of the Fourth Industrial Revolution. Very large data sets can be collected at reasonable costs — for example, they can be captured from sensors, which are widely available, or crowd sourced from the public. However, large data sets with annotation (labelling, structuring, etc.) are very rare, as the annotation process can be very expensive. This is especially the case in biomedical domains, where proper annotation must be done by experts. Therefore, data sets with annotation are mostly small. This poses special challenges in data analysis and addressing such small data challenges needs special methodologies in what can be called small data analytics. There are already some specialist research areas that are related to small data analytics, such as one-shot or zero-shot learning. This Special Issue will cover small data analytics in a broader sense along the following technical challenges.
The first challenge with small data analytics is how to effectively learn with small data sets. Machine learning, especially deep learning, can effectively learn with large data sets. However, they cannot effectively learn with small data sets due to various issues (for example, overfitting, noise, outliers and sampling bias) which can render the learned model ineffective. There are ways to deal with small data sets, such as data augmentation, transfer learning, regularisation and visualisation. However, these methods need skilled people, and their effectiveness is limited.
The second challenge with small data analytics is how to annotate data more easily and cheaply so that large data sets with annotation are not scarce assets anymore. The annotation challenge can be addressed via the use of annotation tools/services in a number of ways. (1) One is to provide annotation tools such as Computer Vision Annotation Tool so that annotation can be done more effectively and easily. (2) One is to outsource the annotation task to an annotation service provider such as Amazon Mechanical Turk. However, none of these approaches actually solves the annotation problem scientifically; rather they do so as a business. To solve the annotation problem scientifically, one desirable approach is to design a new machine learning algorithm that needs minimal feedback from human experts. This will only be possible if domain specific constraints can be imposed in the learning process. This will reduce the model space hence the variance in learning. One research question is thus how to reduce model space by domain specific constraints.
The third challenge is how to leverage available knowledge (common sense, domain specific) implicitly or explicitly in the learning process. When there is a lot of data, machine learning needs a small amount of knowledge (for example, in choosing a machine learning model); when there is not much data, machine learning needs a large amount of knowledge to reduce the model space. It is therefore reasonable to hypothesise that knowledge-based machine learning is a way forward for small data analytics.
This Special Issue will publish papers on small data analytics, including few-shot or zero-shot learning, and those that address the three technical challenges identified above: (1) learning with small datasets (2) annotation in machine learning (3) knowledge-based machine learning. It will also feature contemporary applications where small data is a challenge or where both knowledge and data are needed.
Guest EditorsProf. Hui Wang, Ulster University, UKProf. Ivo Duentsch, Brock University, CanadaProf. Gongde Guo, Fujian Normal University, ChinaProf. Sadiq Ali Khan, University of Karachi, Pakistan
Important DatesDeadline for full paper: 31-May-21First notification: 31-Aug-21Submission of revised manuscript: 15-Oct-21Final notification: 30-Nov-21Submission of final papers: 31-Dec-21Publication Date: 2022