CFG Seminar: A Study on Crowd-AI Interaction towards Fair Human Face Applications

PhD student Ziyi Kou will lead the session, "A Study on Crowd-AI Interaction towards Fair Human Face Applications"

Abstract: Human face images represent a rich set of visual data information that is utilized by various data-driven human facial applications. However, the performance of these applications is usually biased towards the majority demographic group due to the data imbalance issue. To solve the fairness problem, current approaches usually require pre-annotated demographic labels of human face images and focus on developing application-specific machine learning (ML) models that cannot be easily generalized to other ML applications. In this talk, we will discuss the crowd-AI interaction strategies that generate fair human face datasets to improve both the fairness and accuracy performance of human facial applications optimized on such datasets. The talk consists of two case studies. In the first study, we discuss a fair crowdsourcing-based data sampling framework that leverages an efficient batch-level demographic label inference model and a joint fair-accuracy-aware data shuffling method to sample a fair sub-dataset from a large human face dataset with demographic biases. In the second study, we discuss a crowdsourcing-based fair data exchange framework to generate a set of augmented fair face image datasets by leveraging the crowdsourced demographic attribute labels of human face images. Both case studies demonstrate the effectiveness of crowdsourcing strategies to reduce various demographic biases of human face datasets and develop fairness-aware human facial applications.

Readings:
[1] Kou, Z., Zhang, Y., Shang, L., & Wang, D. (2021, June). Faircrowd: Fair human face dataset sampling via batch-level crowdsourcing bias inference. In 2021 IEEE/ACM 29th International Symposium on Quality of Service (IWQOS) (pp. 1-10). IEEE.
[2] Kou, Z., Shang, L., Zeng, H., Zhang, Y., & Wang, D. (2021, December). ExgFair: A Crowdsourcing Data Exchange Approach To Fair Human Face Datasets Augmentation. In 2021 IEEE International Conference on Big Data (Big Data) (pp. 1285-1290). IEEE.

Questions? Contact Lan Li