How does machine learning relate to scientific inquiry loop?
Through machine learning we evaluate plausability of our beliefs on hypotheses given observations (data) via Baye's rule.
Give the setups for supervised, unsupervised and reinforcement learning problems.
Supervised learning setup: given data set $D=\{(x_1, t_1), \ldots, (x_N, t_N) \}$, infer $p(t|x)$. Unsupervised learning setup: given data set $D=\{x_1, \ldots, x_N \} $, infer $p(x)$. Reinforcement learning setup: let $s_t \in S$ be environmental states, $a_t \in A$ be actions and $r_t \colon A \times S \to [0,1]$ be a reward function. Reinforcement objective is to; $$ \begin{align*} \underset{a_t}{\text{maximize}}& &r_t \end{align*} $$