Task Force on Multitask Learning and Multitask Optimization

Task Force Members

Zexuan Zhu, Shenzhen University, China, zhuzx@szu.edu.cn
Liang Feng, Chongqing University, China, liangf@cqu.edu.cn
Ke Tang, University of Science and Technology of China, China, ketang@ustc.edu.cn
Ying-ping Chen, National Chiao Tung University, Taiwan, ypchen@cs.nctu.edu.tw
Qingfu Zhang, City University of Hong Kong, Hong Kong, qingfu.zhang@cityu.edu.hk
Kay Chen Tan, City University of Hong Kong, Hong Kong, kaytan@cityu.edu.hk
Hisao Ishibuchi, Osaka Prefecture University, Japan, hisaoi@cs.osakafu-u.ac.jp
Maoguo Gong, Xidian University, China, gong@ieee.org
Mengjie Zhang, Victoria University of Wellington, New Zealand, Mengjie.Zhang@ecs.vuw.ac.nz
Jacek, Mandziuk, Warsaw University of Technology, Poland, J.Mandziuk@mini.pw.edu.pl
Huynh Thi Thanh Binh, Hanoi University of Science and Technology, Vietnam, binhht@soict.hust.edu.vn
Ahmed Kattan, Ministry of Labor, Saudi Arabia, kattan.ahmed@gmail.com
Jinghui Zhong, South China University of Technology, China, jinghuizhong@gmail.com
Zhun Fan, Shantou University, China, zfan@stu.edu.cn
Amiram Moshaiov, Tel Aviv University, Israel, moshaiov@post.tau.ac.il

Motivation

The human mind possesses the most remarkable ability to manage and execute multiple tasks with apparent simultaneity, e.g., talking while walking. In fact, with the present-day explosion in the variety and volume of incoming information streams that must be absorbed and processed, the opportunity, tendency, and (even) the need to multitask is unprecedented. Therefore, the desirable features of multitasking, particularly with regard to maximal utilization of possibly limited resources, have inspired computational methodologies to tackle multiple tasks at the same time by leveraging the correlations and contradictions across them. As a well-known example, multitask learning has attracted much attention as a subfield of machine learning where multiple learning tasks are performed together, using a shared model representation, such that the relevant information contained in related tasks can be exploited to improve the learning efficiency and generalization performance of task-specific models.

While multi-task learning is largely focused on the augmentation of datasets for improving the prediction quality of machine learning models, it is contended that similar ideas of automatic exploitation of shared knowledge can be directly interpreted in the context of optimization problem-solving for improved decision making. The impact of such a technology is expected to be seen across industrial applications where optimization problems seldom exist in isolation. To this end, multi-task optimization is a newly emerging research area. In the multitasking scenario, solving one optimization problem may assist in solving other optimization problems (i.e., synergetic problem-solving) if these problems bear some form of commonality and/or complementarity in terms of optimal solutions and/or fitness landscapes. Notably, recent advances in cyber-physical systems and the internet of things could rapidly turn optimization into an on-demand service hosted on cloud platforms, such that a variety of optimization tasks would need to be simultaneously executed by the service engine. This will provide a perfect setting for multitasking to harness the underlying synergies between multiple tasks to provide better solutions faster.

Goals

The main goal of this task force is to promote research works on crafting novel algorithms and progressing the theoretical understanding of multitask learning as well as multitask optimization. In particular, we seek the development of truly adaptive techniques that can automatically respond to the level of synergy between tasks. Further, this task force aims at providing a forum for academic and industrial researchers to explore future directions of research that form the bleeding edge of truly intelligent decision making systems.

Scope

The scope of this task force includes the following topics:

• Multitask supervised and unsupervised learning
• Multitask reinforcement learning
• Single/Multi-Objective evolutionary multitasking for continuous and/or combinatorial optimization
• Representations and evolutionary operators customized for multitasking
• Theoretical analyses of evolutionary multitasking
• Evolutionary multitask learning
• Bayesian multitask optimization algorithms
• Handling "many" optimization tasks at once
• Real world applications

Planned Activities

• We are organizing a special issue on Multi-task Evolutionary Computation in the IEEE Transactions on Evolutionary Computation.
  For further details, please see the Call for Papers available at https://cis.ieee.org/images/files/Documents/call-for-papers/tevc/TEVC_SI_MEC_CFP.pdf