DM2 - Representative-based clustering Flashcards

Question 1

What is the primary goal of representative-based clustering?

Accepted Answer

To summarize a cluster using a central point or model so that other points in the cluster are mathematically close to it.

Question 2

What are three common representative-based clustering algorithms?

Accepted Answer

k-means, k-medoids, and Expectation-Maximization (EM).

Question 3

In k-means clustering, how is a cluster's center defined?

Accepted Answer

By the centroid, which is the arithmetic mean of all points assigned to that cluster.

Question 4

What distance metric does the standard k-means algorithm use to minimize variance?

Accepted Answer

Euclidean distance.

Question 5

What is the standard algorithmic approach used to optimize k-means?

Accepted Answer

Lloyd's Algorithm.

Question 6

What is the computational complexity of the standard k-means algorithm?

Accepted Answer

$\mathcal{O}(tkn)$ where $t$ is iterations, $k$ is clusters, and $n$ is data points.

Question 7

How does k-medoids differ from k-means in terms of its cluster center?

Accepted Answer

k-medoids restricts the cluster center (the medoid) to be an actual data point from the dataset.

Question 8

Which distance metric is most commonly associated with k-medoids to improve robustness against outliers?

Accepted Answer

Manhattan distance ($L_{1}$ norm).

Question 9

What is a common heuristic algorithm used to solve k-medoids clustering?

Accepted Answer

Partitioning Around Medoids (PAM).

Question 10

What is the Silhouette Coefficient used for in machine learning?

Accepted Answer

To evaluate the quality of clusters by comparing how similar an object is to its own cluster versus other clusters.

Question 11

How do you interpret a Silhouette score close to 1?

Accepted Answer

The data point is well-matched to its own cluster and poorly matched to neighboring clusters, indicating strong structure.

Question 12

What is the primary advantage of Kernel k-means over standard k-means?

Accepted Answer

It can identify non-linearly separable clusters by using the kernel trick to project data into a higher-dimensional space.

Question 13

What type of clustering does the Expectation-Maximization (EM) algorithm perform when using Gaussian Mixture Models?

Accepted Answer

Soft clustering, where points belong to multiple clusters with varying probabilities.

Question 14

In the EM algorithm, what does the Expectation (E) step compute?

Accepted Answer

The probability (or weight) that each data point belongs to each specific cluster.

Question 15

In the EM algorithm, what does the Maximization (M) step update?

Accepted Answer

The parameters of the probability distributions (such as mean and covariance) based on the weights computed in the E step.

DM2 - Representative-based clustering