Seminar on Algorithms: Clustering
Clustering is a fundamental task in data science that often arises in real-world applications. In the past decades several alternative formulations of the clustering problem have been proposed. These have led to a variety of algorithms, both exact and approximate, using a broad range of optimization techniques such as local search, core-sets, linear programming, etc. Studying the inherent hardness of clustering has led to deep theoretical results, combining insights from geometry and graph theory.
In this seminar/reading group we will study representative and recent results related to clustering, with a focus on algorithmic and complexity aspects. We present/discuss one or two papers each week.
Requirements: ALP3/HA or similar algorithmic background, mathematical maturity.
|k-center greedy approximation and hardness
|NP-hardness of k-means
|impossibility of clustering
|sampling approximation for k-means
|LP-relaxation for median clustering
|graph clustering: cut-based methods
|spectral method for expander-decomposition
|explainable k-means clustering