Multiple-task Learning for Big Data (ML4BD)

Summary and Scope:

Big Data is a term that describes the large volume of data-both structured and unstructured. With the rapid development of new information technologies such as smart phone, mobile game platforms, smart home devices, smart health devices, and wearable computation devices, the amount of created and stored data on global level is almost inconceivable and it just keeps growing. These data is so large and complex that traditional data processing applications are incapable of dealing with them. There are many challenges when addressing big data problem, such as data acquisition, data curation, data storage, data search, data transfer and sharing, data visualization, data query and retrieval, information security, and data analysis (e.g., prediction, user behavior analysis). Big data requires novel data processing techniques to solve some of these challenges jointly, which is related to Multiple-task Learning methodologies.

Multiple-task Learning is a subfield of machine learning in which multiple learning tasks are solved at the same time, while exploiting commonalities and differences across tasks. This can result in improved learning efficiency and prediction accuracy for each task-specific model, when compared to training each of the models separately. Multiple-task Learning is an approach of inductive transfer that improves generalization by using domain information contained in the training signals of related tasks as inductive bias. It does this by learning multiple tasks in parallel using a shared representation, based on the assumption that all tasks can help each other in learning. MTL works because regularization induced by requiring an algorithm to perform well on a related task can be superior to regularization that prevents overfitting by penalizing all complexity uniformly. MTL may be particularly helpful if all related tasks share significant commonalities and are slightly under sampled. With big structured and unstructured data, different tasks on the same data or related data are suitable for MTL framework. Therefore, Multiple-task Learning for Big Data (MTL4BD) have broad applications in many fields, such as online recommendation system, smart home, smart health care, robotics, medical imaging, multimedia application, computer vision, human computer interaction and language processing, etc. We plan to receive about 45 paper submissions from around 30 Universities and research institutions and we will accept about 15 papers plus a review paper totally.

The list of possible topics include, but not limited to:
 Big Data Applications.
o Big data of online recommendation system and business system with multiple tasks: searching, recommendations, ranking, user personalized behavior prediction.
o Big data of natural language processing with multiple tasks: Speech tagging, Segments labeling, Named entity recognition and Semantic labelling.
o Big data of natural imaging processing with multiple tasks: image classifications, image semantic labelling, image searching and scene understanding.
o Big data of human computer interaction with multiple tasks: human computer conversion, vision and audio based human computer interactions and user performance prediction.
o Big data of medical imaging processing with multiple tasks: computer assisted diagnosis including clinical label and clinical score prediction, image registration, feature learning and selection, multiple disease jointly diagnosis and prognosis.

 Multiple-task Learning Methodology
o Shared nodes and layers in Convolutional Deep Neural Network model for multiple tasks.
o Shared feature representations for multiple source data or multiple-task learning.
o Shared parameters in Gaussian Process model, Hidden Markov Chain model on multiple task problems.
o Shared data mean in multiple source data in order to penalize the deviation of each task from the mean task or major task.
o Transfer learning and domain adaption learning among different overlapped tasks or related tasks.
o Multiple task active learning with different activation constraints and categories, constraint-driven active learning across tasks and cost-sensitive active learning across tasks.
o Cluster/graph multiple tasks learning if the multiple tasks are not totally related but have some shared graph or clustered structures.

Submission Guideline

Submissions to this special issue Multiple-task Learning for Big Data (MTL4BD) for possible publications MUST be original and MUST NOT be under consideration for publication in any other journal or conference.

Regarding the submission extended by a previously published workshop or conference paper, it should be explicitly mentioned in the manuscript or the corresponding cover letter. Moreover, the submission must cite the former conference version and technically extend by following the Pattern Recognition Letter rules, in which the technical contribution of the journal version MUST be extended beyond 30%, compared to its conference version.

Authors must follow the formatting and submission instructions of Pattern Recognition Letters at
https://www.journals.elsevier.com/pattern-recognition-letters, and follow the \”Submit Online\” link on that page. Please make sure you mention in your cover letter that you are submitting to this special issue: Multiple-task Learning for Big Data (ML4BD) and please select SI:ML4BD in the submission system.

Important Dates:

 Paper submission period: March 1-31, 2018
 First notification: May 10, 2018
 First revision: June 30, 2017
 Second notification: July 31, 2018
 Second revision: Aug 31, 2018
 Final decision: Sep 30, 2018
 Publication date: Winter 2018 (Tentative)

Guest Editors:
 Dr. Yingying Zhu (Managing Guest Editor), The University of North Carolina at Chapel Hill, United States (zyy@med.unc.edu)
 Dr. Xie Qing, Wuhan University of Technology, China (felixxq@whut.edu.cn)
 Dr. Kimhan Thung, University of Malaya, Malaysia (henrythung@gmail.com)