# Background
Given ***training samples***,
$
\{\mathbf{x}_n,t_n\}
$
- $\mathbf{x}_n$ - training data
- $t_n$ - target values
$
t_n = f(\mathbf{x}_n)
$
find a good approximation to the desired function
$
y(\mathbf{x})\approx f(\mathbf{x})
$
Goal:
- Minimize ***generalization error***
- probability of making an error on *unseen* data points
- cross-validation
## Scenarios
Medical diagnosis
- $\mathbf{x}$ - set of symptoms of a patient
- $f(\mathbf{x})$ - their disease
Property evaluation
- $\mathbf{x}$ - sales information
- $f(\mathbf{x})$ - suggested price
Speech recognition
- $\mathbf{x}$ - recorded signal
- $f(\mathbf{x})$ - sentence in words
# When to Use?
- There is no human expert
- Humans can perform the task but can't explain how
- The desired function changes too frequently
- The application is user-specific
# Methods
- [[Decision Trees]]
- [[Linear Regression]]
- [[K-Nearest Neighbors (kNN)]]
- [[Support Vector Machines (SVM)]]
- [[Neural Networks]]