Training Data | Labeled data (input & output pairs) | Unlabeled data (only input) | Feedback from environment |
Learning Objective | To learn a mapping function from input to output | To discover patterns and relationships in data | To maximize cumulative reward |
Example Use Cases | Image classification, Regression, Speech recognition | Clustering, Dimensionality reduction | Game playing, Robotics |
Teacher/Guide | Provided with correct answers or labels | No explicit teacher or guide | Reward signal from environment |
Model Output | Predictions based on learned patterns | Cluster/group data, Dimension-reduced representation | Actions to take in an environment |
Evaluation Metric | Typically uses metrics like accuracy, loss | Quality of clustering, Reconstruction error | Cumulative reward, Success rate |
Approach Complexity | Often simpler as it has labeled data for direct comparison | More complex as it requires finding patterns in data | Complex due to the interaction with the environment |
Key Challenges | Requires labeled data, may suffer from overfitting | Difficulty in identifying the correct clustering or patterns | Balancing exploration and exploitation |