• an unsupervised clustering algorithm that aims to partition n observations into k clusters in which each observation belongs to the cluster with the nearest mean (or centroid), serving as a prototype of the cluster. -Wikipedia, ASU CSE 575

Is a:

• method

Applications:

• Data Clustering
• Feature Learning
• Recommendation

Advantages:

• very simple
• modifications such as k-means++ compensate for initialization sensitivity

Disadvantages:

• sensitive to initial parameters
• finding the optimal solution is NP-hard