Voronoi Tessellation-based Confidence Decision Boundary Visualization to Enhance Understanding of Active Learning

Jie Chen, Siteng Ma, Dairui Liu, Brian Mac Namee, Ruihai Dong

Keywords: Decision Boundary, Visualization, Active Learning

Abstract:

The current visualizations used in active learning fail to capture the cumulative effect of the model in the active learning process, making it difficult for researchers to effectively observe and analyze the practical performance of different query strategies.
To address this issue, we introduce the confidence decision boundary visualization, which is generated through Voronoi tessellation and evaluated using ridge confidence. This allows better understanding of selection strategies used in active learning. This approach enhances the information content in boundary regions where data distribution is sparse. Based on the confidence decision boundary, we created a series of visualizations to evaluate active learning query strategies. These visualizations capture nuanced variations regarding how different selection strategies perform sampling, the characteristics of points selected by various methods, and the impact of newly sampled points on the model. This enables a much deeper understanding of the underlying mechanisms of existing query strategies.

Primary Area: other topics in machine learning (i.e., none of the above)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 12297