Investigating Bi-Level Optimization for Learning and Vision From a Unified Perspective: A Survey and Beyond

Liu，Risheng1; Gao，Jiaxin2; Zhang，Jin3; Meng，Deyu4; Lin，Zhouchen5

2022-12-01

10.1109/TPAMI.2021.3132674

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE   影响因子和分区

0162-8828

1939-3539

Bi-Level Optimization (BLO) is originated from the area of economic game theory and then introduced into the optimization community. BLO is able to handle problems with a hierarchical structure, involving two levels of optimization tasks, where one task is nested inside the other. In machine learning and computer vision fields, despite the different motivations and mechanisms, a lot of complex problems, such as hyper-parameter optimization, multi-task and meta learning, neural architecture search, adversarial learning and deep reinforcement learning, actually all contain a series of closely related subproblms. In this paper, we first uniformly express these complex learning and vision problems from the perspective of BLO. Then we construct a best-response-based single-level reformulation and establish a unified algorithmic framework to understand and formulate mainstream gradient-based BLO methodologies, covering aspects ranging from fundamental automatic differentiation schemes to various accelerations, simplifications, extensions and their convergence and complexity properties. Last but not least, we discuss the potentials of our unified BLO framework for designing new algorithms and point out some promising directions for future research. A list of important papers discussed in this survey, corresponding codes, and additional resources on BLOs are publicly available at: https://github.com/vis-opt-group/BLO.

Optimization Task analysis Convergence Complexity theory Reinforcement learning Multitasking Computer vision Bi-level optimization Learning and vision applications Value-function-based reformulation Best-response mapping Explicit and implicit gradients

EI ; SCI

英语

其他

National Key R&D Program of China[2020YFB1313503] ; National Natural Science Foundation of China["61731018","61922019","11971220"] ; Shenzhen Science and Technology Program[RCYX20200714114700072] ; Macao Science and Technology Development Fund[061/2020/A2] ; Tianyuan Fund for Mathematics[12026606] ; PKU-Baidu Fund Project[2020BD006]

Computer Science ; Engineering

Computer Science, Artificial Intelligence ; Engineering, Electrical & Electronic

WOS:000880661400105

IEEE COMPUTER SOC

20215111368485

Computer games ; Deep learning ; Game theory ; Job analysis ; Parallel processing systems ; Reinforcement learning

Ergonomics and Human Factors Engineering:461.4 ; Digital Computers and Systems:722.4 ; Artificial Intelligence:723.4 ; Computer Applications:723.5 ; Vision:741.2 ; Probability Theory:922.1

ENGINEERING

Web of Science

1.DUT-RU Intenational School of Information Science and Technology, Dalian University of Technology, 12399 Dalian, Liaoning, China, (e-mail: rsliu@dlut.edu.cn)
2.DUT-RU International School of Information Science Engineering, Dalian University of Technology, 12399 Dalian, Liaoning, China, (e-mail: jiaxinn.gao@outlook.com)
3.Department of Math, Southern University of Science and Technology, 255310 Shenzhen, Guangdong, China, 518055 (e-mail: zhangj9@sustech.edu.cn)
4.Faculty of Science, Institute for Information and System Science, Xi'an, Shaan'xi, China, 710049 (e-mail: dymeng@mail.xjtu.edu.cn)
5.Key Lab. of Machine Perception (MOE), Peking University, Beijing, Beijing, China, 100080 (e-mail: zlin@pku.edu.cn)

Liu，Risheng,Gao，Jiaxin,Zhang，Jin,et al. Investigating Bi-Level Optimization for Learning and Vision From a Unified Perspective: A Survey and Beyond[J]. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,2022,44(12):10045-10067.

Liu，Risheng,Gao，Jiaxin,Zhang，Jin,Meng，Deyu,&Lin，Zhouchen.(2022).Investigating Bi-Level Optimization for Learning and Vision From a Unified Perspective: A Survey and Beyond.IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,44(12),10045-10067.

Liu，Risheng,et al."Investigating Bi-Level Optimization for Learning and Vision From a Unified Perspective: A Survey and Beyond".IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE 44.12(2022):10045-10067.