1. Dimensionality Reduction
   1. Benefits
      1. Computational ease
      2. Less overfitting
   2. Techniques
      1. Feature Selection
         1. Criteria
            1. Correlation among features
            2. Feature Discriminability
         2. Techniques
            1. Forward/Backward searches
            2. Branch and bound
      2. Feature Extraction
         1. Criteria
            1. align/deform/unfold data to be discriminable in few dimensions
         2. Techniques
            1. Assumptions
2. Classification
3. Machine
   1. Finite State Automaton (FSA)
   2. Quantum Computing (=FSA?)
   3. Human?--no just machines :)
4. Learning
   1. What?
      1. Output given input
   2. Why?
      1. Automation
         1. Proper utilization of human resource
            1. Delegate unpleasant jobs (produce output given input) to non-humans :)
         2. less errors
         3. fast
   3. When?
      1. when prediction works
   4. How?
      1. Input Processing
         1. What?
            1. Transform raw data into data having particular characteristics
         2. why?
            1. Many prediction functions share certain requirements from input
            2. Representation
            3. Structure
            4. Point pattern
            5. Vector
            6. Values/ Data Types
            7. Nominal
            8. Ordinal
            9. Interval
            10. Ratio
            11. Noise Model
            12. Additive noise
            13. Missing Data
            14. Size
         3. How?
            1. Handling Missing Data
            2. Dimensionality Reduction
            3. Changing data type
            4. Changing data structure
      2. Training the Prediction Function
         1. Assumptions
            1. Function class restriction
            2. why?
            3. Accuracy
            4. How?
            5. By design
            6. Choose linear or quadratic prediction function
            7. Regularization
            8. Input domain restriction
            9. Structure restriction
            10. Data type restriction
            11. Noise restriction
         2. Techniques
            1. Output type
            2. Classification
            3. Regression
            4. Training data
            5. Input-output pairs
            6. Only Input
            7. Link/ no-link constraints
            8. Availability
            9. Function
            10. What is modeled?
            11. Data generation function: Generative
            12. Class separating boundary: Discriminative
            13. Function combining outputs of existing prediction schemes: Meta-learners
            14. Complexity
            15. Temporal consistency
            16. Drifting objective learning
      3. Testing the Prediction Function