# Week4 - What can be learned Week 4 Learning Classifications...

This preview shows pages 1–4. Sign up to view the full content.

Week 4 Learning Chapter 18, 20 What is learning? ! Improving performance based on experience: ! Discovering new relationships between input and output Which input data is important and which can be disregarded ! Discovering properties of the environment Learning the configuration of a maze ! Describing experience in a concise way ! Which features are best mapping from input to output ! Better than lookup table with previous experiences Learning Agents ! Agents have two major components: ! Performance element: Selecting action to take at each timestep ! Learning element: Improving the selection process by comparing outcome to optimal outcome ! Performance element must be written so that it can be modified by learning ! Logical statements or very modular code What can be learned? ! Classifications: ! Identify hand-written digits ! Filter mail into spam/not spam ! Find the face in a photo ! Actions: ! Robot balances upright on two legs ! Autopilot flies level ! Keep vehicle in lane Problem generator suggests exploratory actions rather than just letting the performance element select the action that it has already learned is best Feedback ! Indicates the results of agent’s output ! Can be in terms of correct answer provided by a friendly teacher ! Supervised learning ! Agent compares its answer to answer key ! Good for precise answers like classifications (recognizing handwritten digits) ! Can be a system of rewards and punishments ! reinforcement ! Better for gradient actions (balancing on two legs)

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Prior Knowledge ! Learning can begin with the first experience or can benefit from built-in bias ! Canned chess moves ! Ratio of letter occurrences in English Inductive Learning ! A form of supervised learning ! Hypothesizes a function mapping inputs to correct outputs ! Presented with example Tuples (x,f(x)): ! x is input ! f(x) is output of function applied to x Hypotheses ! A hypothesis h is an approximation of the true function f that you are trying to learn ! Pure inductive inference ! Given {(x,f(x))}, return h(x) which approximates f(x) ! The space of all hypothesis functions is H ! This is chosen by the person designing the learning algorithm Inductive learning x f(x) Inductive learning H 1 : first degree polynomials (lines) Inductive learning H 2 : second degree (ax 2 +bx+c)
Inductive learning H 3 : third degree (ax 3 +bx 2 +cx+d) Inductive learning H 4 : fourth degree Inductive learning H 12 : twelfth degree Inductive learning Which is the appropriate degree?

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### Page1 / 10

Week4 - What can be learned Week 4 Learning Classifications...

This preview shows document pages 1 - 4. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online