Ciolacu, Gabriela; Jin, David

Requirements:

• English Language preferred.
• Basic Knowledge of Machine Learning (ML) is appreciated.

Research Problem:

• With the increasing usage of machine learning (ML) as a crucial part of our daily lives, the question of fairness, notably algorithmic fairness, has become a prominent point of discussion in society and research. Algorithmic fairness in centralized ML systems has significantly progressed in the past decades. However, with the shift towards decentralized systems, fairness must be rethought.
• The importance of Collaborative Distributed Machine Learning (CDML) is undeniable. Decentralized and CDML address data and computing power scarcity arising in centralized ML systems, and it is increasingly used in health, finance, and IoT systems.
• However, the lenses of algorithmic fairness do not suffice to understand the subtleties of CDML, as algorithmic fairness has been predominantly discussed in centralized ML settings and centers around statistical bias. In decentralized systems such as multi-agent settings, fairness includes resource allocations (e.g., training data and computing resources, static and dynamic fairness), political and legal requirements (e.g., different countries; different power distribution that impose the limits to fairness), the need to reciprocate in collaboration and the associated risk of a free rider.
• To ensure a granular understanding of fairness in CDML systems, we address the following question: How do the prevailing perspectives in CDML fairness differ from centralized ML fairness?

Objectives:

• Objective 1: Describe prevailing perspectives in CDML fairness through a structured literature review.
• Objective 2: Describe prevailing perspectives in centralized ML fairness based on a few selected papers. 
• Objective 3: Compare the derived perspectives of CDML and centralized ML.

Method:

Literature Review.

Expected Contributions:

• Contribution to research: We extend the literature on fairness by offering a granular overview of fairness in centralized and decentralized ML systems. 
• Contribution to practice: We inform practice to design fair (CD)ML systems by describing the prevailing perspectives in (CD)ML and comparing them.

Introductory Literature:

• Caton, S., & Haas, C. (2020). Fairness in machine learning: A survey. ACM Computing Surveys.
• Dolata, M., Feuerriegel, S., & Schwabe, G. (2022). A sociotechnical view of algorithmic fairness. Information Systems Journal, 32(4), 754-818.
• Jin, D., Kannengießer, N., Rank, S., & Sunyaev, A. (2023). A Design Toolbox for the Development of Collaborative Distributed Machine Learning Systems. arXiv preprint arXiv:2309.16584.
• Lyu, L., Xu, X., Wang, Q., & Yu, H. (2020). Collaborative fairness in federated learning. Federated Learning: Privacy and Incentive, 189-204.
• Mitchell, S., Potash, E., Barocas, S., D'Amour, A., & Lum, K. (2021). Algorithmic fairness: Choices, assumptions, and definitions. Annual Review of Statistics and Its Application, 8, 141-163.
• Pessach, D., & Shmueli, E. (2022). A review on fairness in machine learning. ACM Computing Surveys (CSUR), 55(3), 1-44.
• Yang, Q., Liu, Y., Chen, T., & Tong, Y. (2019). Federated machine learning: Concept and applications. ACM Transactions on Intelligent Systems and Technology (TIST), 10(2), 1-19.