240. An Exploratory Study on Clustering Models under Perturbations Based on a Bilevel Optimization Framework
Invited abstract in session MC-7: Bilevel Optimization in Data Science, stream Bilevel and multilevel optimization.
Monday, 14:00-16:00Room: B100/5015
Authors (first author is the speaker)
| 1. | Yutong Zheng
|
| Beijing Institute of Technology |
Abstract
Clustering, as one of the core tasks in unsupervised learning, plays a crucial and indispensable role in numerous disciplines and application domains. However, the robustness of clustering is increasingly threatened by noisy data. This presentation provides an overview of our preliminary investigation into the impact of perturbations on clustering results. Using a bilevel optimization framework, we have studied a perturbation bound that ensures cluster stability, thereby establishing a connection to adversarial data attacks. Furthermore, we examine a metric for evaluating the degree of change in clustering results and derive its explicit expression and monotonicity properties. Future research will focus on extending the framework to higher dimensions and larger scales, while further exploring perturbation robustness and adversarial data attacks, thus addressing real-world large-scale challenges.
Keywords
- Large-scale optimization
- Linear and nonlinear optimization
- Second- and higher-order optimization
Status: accepted
Back to the list of papers