94
STICK-BREAKING CONSTRUCTIONS Patrick Dallaire June 10 th , 2011

Stick-Breaking Constructions

  • Upload
    huslu

  • View
    31

  • Download
    0

Embed Size (px)

DESCRIPTION

Stick-Breaking Constructions. Patrick Dallaire June 10 th , 2011. Outline. Introduction of the Stick-Breaking process. Outline. Introduction of the Stick-Breaking process Presentation of fundamental representation. Outline. Introduction of the Stick-Breaking process - PowerPoint PPT Presentation

Citation preview

Page 1: Stick-Breaking Constructions

STICK-BREAKING CONSTRUCTIONSPatrick DallaireJune 10th, 2011

Page 2: Stick-Breaking Constructions

Outline Introduction of the Stick-Breaking

process

Page 3: Stick-Breaking Constructions

Outline Introduction of the Stick-Breaking

process Presentation of fundamental

representation

Page 4: Stick-Breaking Constructions

Outline Introduction of the Stick-Breaking

process Presentation of fundamental

representation The Dirichlet process The Pitman-Yor process The Indian buffet process

Page 5: Stick-Breaking Constructions

Outline Introduction of the Stick-Breaking

process Presentation of fundamental

representation The Dirichlet process The Pitman-Yor process The Indian buffet process

Definition of the Beta process

Page 6: Stick-Breaking Constructions

Outline Introduction of the Stick-Breaking

process Presentation of fundamental

representation The Dirichlet process The Pitman-Yor process The Indian buffet process

Definition of the Beta process A Stick-Breaking construction of Beta

process

Page 7: Stick-Breaking Constructions

Outline Introduction of the Stick-Breaking

process Presentation of fundamental

representation The Dirichlet process The Pitman-Yor process The Indian buffet process

Definition of the Beta process A Stick-Breaking construction of Beta

process Conclusion and current work

Page 8: Stick-Breaking Constructions

The Stick-Breaking process

Page 9: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length

Page 10: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length

Page 11: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 12: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 13: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 14: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 15: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 16: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 17: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 18: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 19: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 20: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 21: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining

stick is broken by sampling the proportion to cut

Page 22: Stick-Breaking Constructions

The Stick-Breaking process

Assume a stick of unit length At each iteration, a part of the remaining stick

is broken by sampling the proportion to cut How should we sample these proportions?

Page 23: Stick-Breaking Constructions

Beta random proportions Let be the proportion to cut at

iteration

Page 24: Stick-Breaking Constructions

Beta random proportions Let be the proportion to cut at

iteration The remaining length can be expressed

as

Page 25: Stick-Breaking Constructions

Beta random proportions Let be the proportion to cut at

iteration The remaining length can be expressed

as

Thus, the broken part is defined by

Page 26: Stick-Breaking Constructions

Beta random proportions Let be the proportion to cut at

iteration The remaining length can be expressed

as

Thus, the broken part is defined by

We first consider the case where

Page 27: Stick-Breaking Constructions

Beta distribution The Beta distribution is a density

function on

Parameters and control its shape

Page 28: Stick-Breaking Constructions

The Dirichlet process

Page 29: Stick-Breaking Constructions

The Dirichlet process Dirichlet processes are often used to

produce infinite mixture models

Page 30: Stick-Breaking Constructions

The Dirichlet process Dirichlet processes are often used to

produce infinite mixture models Each observation belongs to one of the

infinitely many components

Page 31: Stick-Breaking Constructions

The Dirichlet process Dirichlet processes are often used to

produce infinite mixture models Each observation belongs to one of the

infinitely many components The model ensures that only a finite

number of components have appreciable weight

Page 32: Stick-Breaking Constructions

The Dirichlet process A Dirichlet process, , can be constructed

according to a Stick-Breaking process

Where is the base distribution and is a unit mass at

Page 33: Stick-Breaking Constructions

Construction demo

Page 34: Stick-Breaking Constructions

Construction demo

Page 35: Stick-Breaking Constructions

Construction demo

Page 36: Stick-Breaking Constructions

Construction demo

Page 37: Stick-Breaking Constructions

Construction demo

Page 38: Stick-Breaking Constructions

Construction demo

Page 39: Stick-Breaking Constructions

Construction demo

Page 40: Stick-Breaking Constructions

Construction demo

Page 41: Stick-Breaking Constructions

Construction demo

Page 42: Stick-Breaking Constructions

Construction demo

Page 43: Stick-Breaking Constructions

Construction demo

Page 44: Stick-Breaking Constructions

Construction demo

Page 45: Stick-Breaking Constructions

Construction demo

Page 46: Stick-Breaking Constructions

Construction demo

Page 47: Stick-Breaking Constructions

Construction demo

Page 48: Stick-Breaking Constructions

Construction demo

Page 49: Stick-Breaking Constructions

The Pitman-Yor process

Page 50: Stick-Breaking Constructions

The Pitman-Yor process A Pitman-Yor process, , can be

constructed according to a Stick-Breaking process

Where and

Page 51: Stick-Breaking Constructions

Evolution of the Beta cuts The parameter controls the speed at

which the Beta distribution changes

Page 52: Stick-Breaking Constructions

Evolution of the Beta cuts The parameter controls the speed at

which the Beta distribution changes The parameter determines initial

shapes of the Beta distribution

Page 53: Stick-Breaking Constructions

Evolution of the Beta cuts The parameter controls the speed at

which the Beta distribution changes The parameter determines initial

shapes of the Beta distribution When , there is no changes over

time and its called a Dirichlet process

Page 54: Stick-Breaking Constructions

Evolution of the Beta cuts The parameter controls the speed at

which the Beta distribution changes The parameter determines initial

shapes of the Beta distribution When , there is no changes over

time and its called a Dirichlet process

MATLAB DEMO

Page 55: Stick-Breaking Constructions

The Indian Buffet process

Page 56: Stick-Breaking Constructions

The Indian Buffet process The Indian Buffet process was initially

used to represent latent features

Page 57: Stick-Breaking Constructions

The Indian Buffet process The Indian Buffet process was initially

used to represent latent features Observations are generated according to

a set of unknown hidden features

Page 58: Stick-Breaking Constructions

The Indian Buffet process The Indian Buffet process was initially

used to represent latent features Observations are generated according to

a set of unknown hidden features The model ensure that only a finite

number of features have appreciable probability

Page 59: Stick-Breaking Constructions

The Indian Buffet process

Recall the basic Stick-Breaking process

Page 60: Stick-Breaking Constructions

The Indian Buffet process

Recall the basic Stick-Breaking process

Page 61: Stick-Breaking Constructions

The Indian Buffet process

Recall the basic Stick-Breaking process Here, we only consider the remaining

parts

Page 62: Stick-Breaking Constructions

The Indian Buffet process

Recall the basic Stick-Breaking process Here, we only consider the remaining

parts

Page 63: Stick-Breaking Constructions

The Indian Buffet process

Recall the basic Stick-Breaking process Here, we only consider the remaining

parts Each value corresponds to a feature

probability of appearance

Page 64: Stick-Breaking Constructions

Summary

Page 65: Stick-Breaking Constructions

Summary The Dirichlet process induces a

probability over infinitely many classes

Page 66: Stick-Breaking Constructions

Summary The Dirichlet process induces a

probability over infinitely many classes This is the underlying de Finetti mixing

distribution of the Chinese restaurant process

Page 67: Stick-Breaking Constructions

De Finetti theorem It states that the distribution of any

infinitely exchangeable sequence can be written

where is the de Finetti mixing distribution

Page 68: Stick-Breaking Constructions

Summary The Dirichlet process induces a

probability over infinitely many classes This is the underlying de Finetti mixing

distribution of the Chinese restaurant process

The Indian Buffet process induces a probability over infinitely many features

Page 69: Stick-Breaking Constructions

Summary The Dirichlet process induces a

probability over infinitely many classes This is the underlying de Finetti mixing

distribution of the Chinese restaurant process

The Indian Buffet process induces a probability over infinitely many features

Its underlying de Finetti mixing distribution is the Beta process

Page 70: Stick-Breaking Constructions

The Beta process

Page 71: Stick-Breaking Constructions

The Beta process This process

Page 72: Stick-Breaking Constructions

Beta with Stick-Breaking The Beta distribution has a Stick-

Breaking representation which allows to sample from

Page 73: Stick-Breaking Constructions

Beta with Stick-Breaking The Beta distribution has a Stick-

Breaking representation which allows to sample from

The construction is

Page 74: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 75: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 76: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 77: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 78: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 79: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 80: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 81: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 82: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 83: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 84: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 85: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 86: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 87: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 88: Stick-Breaking Constructions

Beta with Stick-Breaking

Page 89: Stick-Breaking Constructions

Beta with Stick-Breaking The Beta distribution has a Stick-

Breaking representation which allows to sample from

The construction is

Page 90: Stick-Breaking Constructions

The Beta process A Beta process is defined as

as , and is a Beta process

Page 91: Stick-Breaking Constructions

Stick-Breaking the Beta process

The Stick-Breaking construction of the Beta process is such that

Page 92: Stick-Breaking Constructions

Stick-Breaking the Beta process

Expending the first terms

Page 93: Stick-Breaking Constructions

Conclusion We briefly described various Stick-Breaking

constructions for Bayesian nonparametric priors

These constructions help to understand the properties of each process

It also unveils connections among existing priors

The Stick-Breaking process might help to construct new priors

Page 94: Stick-Breaking Constructions

Current work Applying a Stick-Breaking process to

select the number of support points in a Gaussian process

Defining a stochastic process for unbounded random directed acyclic graph

Finding its underlying Stick-Breaking representation