Machine Learning - created from Mind Map

Page 1

Supervised Learning

Vapnik-Chervonenkis (VC) Dimension

Classification

Regression

Modeling

Bayesian Decision Theory

P(C|x) = P(C) * p(x|C) / p(x)

posterior p(C | x)

= prior * likelihood / evidence

Risk function

Reject class

Discriminant Functions g(x)

Associative Rule

prior P(C = i)

probability of being class i

likelihood p(x|C)

conditional probability,

probability of event x given being class C

evidence p(x)

marginal probability

probabilty of event x

Support(X,Y) ≡ P(X,Y)

Confidence(X → Y) ≡ P(Y|X)

Lift(X → Y) = P(X,Y) / (P(X) * P(Y))

= P(Y|X) / P(Y)

kind of descripting relationship btw X and Y

Triple tradeoff

Math notation

Complexity

[Underfitting]

Too low complexity, it may give high probability of false positive.

amount of data points

Generalisation error

[Overfitting]

Hypothesis class is too complicated and fit the training data points.

However, the new data points may match the hypothesis and the classification error may raise.

g(x|θ)

where g(·) is the model, x is the input, and θ are the parameters

Loss function, L(·)

E(θ|X) =

θ∗ =argminE(θ|X)