Unsupervised Learning Concepts

Unsupervised Learning

• Unsupervised learning uses unlabeled training data
• The system learns without a teacher
• Training data is unlabeled (Figure 1-7)

Unsupervised Tasks

• Clustering: Machine learning tasks to find patterns of similarity in data samples and cluster them into groups based on shared attributes/features.

• Visualization: Algorithms converting complex, unlabeled data into 2D or 3D representations. These algorithms aim to preserve data structure as much as possible

• Dimensionality Reduction: Simplifying data by merging correlated features into one. Useful to reduce data dimensionality and enhance speed of other algorithms

• Anomaly Detection: Identifying unusual data instances (outliers) to detect anomalies like fraud or manufacturing defects.

• Association Rule Mining: Identifying relationships between attributes in large datasets. An example would be finding that customers purchasing barbecue sauce and potato chips tend also buy steak.

Clustering Techniques

• K-Means: A common clustering algorithm that groups data points into clusters.
• DBSCAN: A clustering algorithm that automatically determines the number of clusters in data.
• Hierarchical Cluster Analysis (HCA): A hierarchical clustering algorithm that clusters data points into a tree-like structure.

Anomaly Detection Techniques

• Nearest Neighbors
• Discriminant Analysis

Association Rule Techniques

• Apriori: Algorithm to discover association rules.
• Eclat: Algorithm to discover association rules.

Semi-Supervised Learning

• Some algorithms can deal with partially labeled training data
• This typically involves a lot of unlabeled data and a bit of labeled data.

Reinforcement Learning

• A learning system (an agent) observes the environment.
• It selects actions and gets rewards in return.
• The agent learns the best strategy (policy) to maximize rewards over time.

Learning Types

• Batch Learning: The system is incapable of learning incrementally and needs all available data to train. Training is done offline and the system operates without any further learning.
• Online Learning: The system trains incrementally via feeding data sequentially. Training is fast and cheap, as it accommodates new data flow.

Model-Based vs Instance-Based Learning

• Instance-Based Learning: The system learns from data by heart and generalizes/predicts based on similarities.
• Model-Based Learning: The system creates a model from data (e.g., an equation) with parameters that make predictions.

Test your knowledge of unsupervised learning techniques and tasks, including clustering, visualization, and anomaly detection. This quiz covers essential methods that enable machines to learn from unlabeled data, revealing patterns and relationships within the data.