| Task | Predicts output labels for input data | Finds patterns and relationships in input data | Learns to make decisions through trial and error |
| Input-Output Data | Requires labeled data (input-output pairs) | Works with unlabeled data (no output labels) | Interacts with an environment through actions |
| Examples | Image classification, sentiment analysis, regression | Clustering, dimensionality reduction, anomaly detection | Game playing, robot control, autonomous driving |
| Training Process | Algorithm learns from labeled data with known outcomes | Algorithm learns patterns from unlabeled data | Learns from feedback in the form of rewards/punishments |
| Evaluation Metrics | Accuracy, precision, recall, F1-score | Internal metrics like clustering quality, silhouette score | Cumulative rewards, success rate, convergence speed |
| Model Interpretation | Provides insights into decision-making processes | Limited interpretability due to lack of labeled data | Often complex and difficult to interpret |
| Common Algorithms | Linear Regression, Decision Trees, SVM, Neural Networks | K-Means, DBSCAN, Principal Component Analysis (PCA) | Q-Learning, Deep Q Networks (DQN), Policy Gradient |
| Use Cases | Predictive modeling, classification tasks | Data exploration, feature learning, anomaly detection | Robotics, gaming, autonomous systems |
| Feedback | Requires correct outputs for training | No explicit feedback required | Feedback in the form of rewards or penalties |
| Example Formula | Output = f(input_features) | Clustering or reduced representation of input data | Action = f(state) |
