jeffreys' and bdeu priors for model selection

24
Jeffreys' and BDeu Priors for Model Selection WITMSE 2016 Helsinki, Finland, September 20 Joe Suzuki (prof-joe) Joe Suzuki (Osaka Univ., Japan)

Upload: joe-suzuki

Post on 29-Jan-2018

231 views

Category:

Science


2 download

TRANSCRIPT

Page 1: Jeffreys' and BDeu Priors for Model Selection

Jeffreys' and BDeu Priors for Model Selection

WITMSE 2016

Helsinki, Finland, September 20Joe Suzuki(prof-joe)

Joe Suzuki (Osaka Univ., Japan)

Page 2: Jeffreys' and BDeu Priors for Model Selection

Goal and Contributions

[Goal] Compare for model selection

• BDeu (Bayesian Dirichlet equivalent uniform)

• Jeffreys prior (T-K estimator)

[Contribution]

Mathematically Proves

Page 3: Jeffreys' and BDeu Priors for Model Selection

Road Map

1. Bayesian Dirichlet Scores

2. BDeu and Jeffreys Scores

3. A Found Property and its Proof

4. Main Theorem

5. Regularity in Model Selection

6. Summary

Page 4: Jeffreys' and BDeu Priors for Model Selection

Assign a Prob. to each Seq.

Page 5: Jeffreys' and BDeu Priors for Model Selection

Express a Prob. by the product of Cond. Probs.

Page 6: Jeffreys' and BDeu Priors for Model Selection

Simultaneous Probs.

Page 7: Jeffreys' and BDeu Priors for Model Selection

Cond. Probs.

Page 8: Jeffreys' and BDeu Priors for Model Selection

BDeu and Jeffreys’ Prior

Page 9: Jeffreys' and BDeu Priors for Model Selection
Page 10: Jeffreys' and BDeu Priors for Model Selection

Example 1 : Bayesian Network Structure Learning (BNSL)

Page 11: Jeffreys' and BDeu Priors for Model Selection

Example 2: Independence Testing

Page 12: Jeffreys' and BDeu Priors for Model Selection

A Motivating Example

Page 13: Jeffreys' and BDeu Priors for Model Selection

A Found Property

Page 14: Jeffreys' and BDeu Priors for Model Selection

Sketch of J(n)>0 for BDeu

Page 15: Jeffreys' and BDeu Priors for Model Selection

Sketch of J(n)≦0 for Jeffreys’

Page 16: Jeffreys' and BDeu Priors for Model Selection

An Intuitive Reasoning

Page 17: Jeffreys' and BDeu Priors for Model Selection

Main Theorem

Page 18: Jeffreys' and BDeu Priors for Model Selection

Examples

more likely

unlikely

Page 19: Jeffreys' and BDeu Priors for Model Selection

Regularity in Model Selection

Fitness + Simplicity → optimal

(-1) x Likelihood + Penalty Term → min

Newton’sLaw of Motion

MaxwellEquations

If model A is better than model B w.r.t. fitness and simplicity,model A should be chosen (regularity).

Information CriteriaLASSO

Page 20: Jeffreys' and BDeu Priors for Model Selection

BDeu violates regularity in model selection

Z XZ X

Y

Y X

Page 21: Jeffreys' and BDeu Priors for Model Selection

B&B for efficient BNSL (Depth First Search)

Page 22: Jeffreys' and BDeu Priors for Model Selection

Those bounds utilize regularity

Campos and Ji 2011 figured out one (=nice)

but the bound is not efficient (experiments).

Designing Pruning rules for BDeu is HARDer.

because regularity cannot be assumed

Page 23: Jeffreys' and BDeu Priors for Model Selection

Bayes Prior

Based on his/her Belief:

Nobody should reject it from a general point of view.

BDeu violates regularity

contradicts with Newton, Maxwell, Information Critreria, LASSO, etc.

People might notice that their beliefs have been wrong, after knowing the new result in this paper.

Page 24: Jeffreys' and BDeu Priors for Model Selection

Summary

The prior behind BDeu might have been based on a wrong belief That contradicts regularity in model selection

Future Work: Consider NML and others in a similar way