07.1 Machine Learning

Machine Learning

field of study in artificial learning

Create a model

ML, main task is to _ _ _ for future behavior given past data or experience

Understand the structure of data

ML, goal (1)

Computational Models

ML, then fit the data into _ _, goal (2)

Prediction

predict future events using models derived from past data, why learn? (1)

Diagnostics

know the most probable causes or reasons of certain events, why learn? (2)

Summarization

combine contents of different data into a concise form, why learn? (3)

Parameters

e.g., probabilities in a Bayesian network, or utility functions indicating desirability of states, what to learn? (1)

Structures

e.g., relationship among variables in a model, what to learn? (2)

Hidden Concepts

new information that might be inferred through observing data or human behavior which helps on making sense of data, what to learn? (3)

Passive

agents learn by observing

Active

agents learn by interacting with the environment

Online

agents learn while receiving data

Offline

agents starts learning once all data is received

Generative

models the actual distribution of each classes of data, models of learning (1)

Generative

learns the joint probability distribution, models of learning (1)

Discriminative

models the decision boundary between classes of data, models of learning (2)

Discriminative

learns the conditional probability distribution

Supervised

agent is given input-output pairs and learns a function that maps from input to output, types of learning (1)

Classification, Regression

two examples under supervised

Unsupervised

agent learns patterns in the input even though no explicit output is supplied

Clustering, Dimensionality reduction

two examples under unsupervised

Reinforcement

agent learns from a series of reinforcement

Reinforcement

does not require training data nor supervision, only interaction with the environment on top of its own decision making policy

Data

_ drives everything, importance of data