Lecture1-4Components of ML

Components of ML: Metaphor using credit approval

The bank has the applicant information, and there is unknown pattern to be learned, which is used to solve the final question:
whether approving credit card good is good for bank?

Basic Notations

• input:$x \in \mathcal{X}$ (custorer application)
• output:$y\in \mathcal{Y}$ (good/bad after approving credit card)
• unknown pattern to be learned $\Longleftrightarrow$ target function:
  $f$: $\mathcal{X} \rightarrow \mathcal{Y}$ (ideal credit approval formula)
• data $\Longleftrightarrow$ training examples: $\mathcal{D}=\{(\boldsymbol{x_1}, y_1), (\boldsymbol{x_2}, y_2), ... , (\boldsymbol{x_N}, y_N)\}$ (historical records in bank)
• hypothesis $\Longleftrightarrow$ skill with hopefully good performance:
  $g$: $\mathcal{X} \rightarrow \mathcal{Y}$ (‘learned’ formula to be used)

$\{(\boldsymbol{x_N}, y_N)\}$ from $f \rightarrow$ ML $\rightarrow g$

$f$ is unkown and $g$ is learned!! Hopefully $g \approx f$, and perfection is impossible for unknown $f$.

Learning Flow1

There will be many $h$ in the hypothesis set $\mathcal{H}$, and then we hope to select one good $h$ as $g$.

Hypothesis set $\mathcal{H}$
- can contain good and bad hypotheses
- up to $\mathcal{A}$ to pick the ‘best’ $h$ as $g$

$\mathcal{A}$ is the ML algorithm.

Learning Model = $\mathcal{A}$ and $\mathcal{H}$

Learning Flow 2