...
This presentation is the property of its rightful owner.
Sponsored Links
1 / 29

ขั้นตอนวิธีเชิงพันธุกรรมสำหรับการอนุมานเครื่องจักรสถานะจำกัด PowerPoint PPT Presentation


  • 58 Views
  • Uploaded on
  • Presentation posted in: General

ขั้นตอนวิธีเชิงพันธุกรรมสำหรับการอนุมานเครื่องจักรสถานะจำกัด. อาจารย์ที่ปรึกษาวิทยานิพนธ์ รศ. ดร. ประภาส จงสถิตย์วัฒนา ประธานกรรมการ ศ. ดร. ชิดชนก เหลือสินทรัพย์ กรรมการ ผศ. ดร. บุญเสริม กิจศิริกุล ดร. ณชล ไชยรัตนะ เสนอโดย นายนัทที นิภานันท์ เลขประจำตัว 403 02410 21.

Download Presentation

ขั้นตอนวิธีเชิงพันธุกรรมสำหรับการอนุมานเครื่องจักรสถานะจำกัด

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

Presentation Transcript


4059834

. .

. .

. .

.

403 02410 21


The story so far

The Story so far

  • There is a task X

  • Process A is a better way to do task X than any previously known method

    • under some measurement

  • A can be improved

  • Finding method A1 which is better than A


Presentation outline

Presentation Outline

  • What is the task X?

  • What is process A?

    • and B, C, etc.

    • Why A is better than B, C, etc.?

  • What point in A that can be improved

  • Boring stuffs (but essential)

    • work plan, objective, scopes, benefit


Introduction

Introduction

Target Machines

HypothesisMachine

? ? ?

  • Mimic the target machine

INPUT

OUTPUT

LearningMethod


Introduction1

Introduction

InductiveInference

Process of hypothesizing a general rule from example

...

GrammaticalInference

Inference of any structure that can recognize a language

DFA Inference

...

Inference of DFA


Application

Application

  • Digital circuit design

    • synthesis of finite state controller from observed I/O signal


Related works

Related Works

GrammaticalInference

PDA

TuringMachine

DFA

Heuristic

Minimal Inference

  • TraxBar

  • EDSM

  • Blue-fringe

Method A

Search

GA

  • Biermann

  • BIC

  • Aporntewan


Heuristic method characteristic

Heuristic Method : characteristic

  • Fast, highly scalable

  • No constraint on the size of hypothesis

  • O(T3H)


Search method characteristic

Search Method : characteristic

  • Slower than state heuristic

  • Very strong constraint on the size of hypothesis

  • Better accuracy than heuristic when training set is sparse

  • Search space is exponential on the size of training set (on fixed target size)

    • O(HT)


Ga method characteristic

GA Method : characteristic

  • Slow

  • Strong constraint on the size of hypothesis

  • Search space is constant on the size of training set (on fixed target size)


Method choosing

Method Choosing

SizeConstrain?

Blue-fringe

LargeTraining set

BIC

GA


Heuristic method

Heuristic Method

  • State merging algorithm

    • Construct a prefix tree acceptor from given examples

    • Merge a pair of states

0

C

Positive Example

00

1

Negative Example

10

B

0

A

0

E

D

1


Heuristic method cont

Heuristic Method (cont.)

D

G

  • Each merge introduce new constrain

  • Early merge should be correct

B

E

H

A

C

F

I

D

G

B

A

C

E F

H I


Heuristic method variation

Heuristic Method : variation

  • TraxBar algorithm

    • Merge by Breadth First Search order

  • EDSM algorithm

    • Merge by score of evident

    • Compute score on every pairs

  • Blue-fringe algorithm

    • Merge by score of evident

    • Compute score only in candidate pairs

      • Much faster than EDSM, with very little accuracy loss


Heuristic method blue fringe

Heuristic Method : Blue-fringe

  • Starting with red at root

  • Children of red is blue

  • Compute and merge only red-blue pair

  • blue can be promoted to red


Search method

Search Method

  • Based on Biermanns algorithm

  • Create Loop Free DFA L = (Q,,,,,q0)

  • Find mapping function F(q) of the states of L to the states q of hypothesis DFA M = (Q,,,,,q0)

    • another form of state merging

    • use exhaustive searching

  • Define Si as the index of the state in the target automaton which state qi in the LFDFA maps to. F(qi) = qSi


Search method cont

Search Method (cont.)

  • Search step (assume hypothesis of N states)

    • 1. Select variable Sj to be assigned from unassigned S

    • 2. Assign value from 0 .. N-1 to Sj, if no more value exists, undo last assignment.

    • 3. If current assignment conflict with the constraints, undo and go to step 2. Else go to step 1.


Search method cont1

Search Method (cont.)

  • Training set pose constraint on S

    • incompatible state

  • Problem can be viewed as constraint satisfactory problem (CSP)


Search method bic

Search Method : BIC

  • By Oliveira and Silva

  • Specialized CSP solver

    • Conflict diagnosis

      • analyze of conflict

      • remember the conflict for future prunning

    • Non-chronological backtracking

      • backjump to the level of the cause of conflict


Ga method

GA Method

  • Search along all less than or equal n-states Mealy machine

    • impose target size constraint

  • Evaluate according to consistency of training set

    • Larger training set does not expand the search space

      • but took (linearly) more time in evaluation


Ga method aporntewan s method

GA Method: Aporntewans Method

  • Encode and in bit string

  • Single point crossover

  • Evaluate by counting different output bit

...

Next State

Output

Next State

Output

Next State

Output

Next State

Output

0-transition

1-transition

0-transition

1-transition

State 0

State N

HypothesisMachine

HypothesisOUTPUT

INPUTSequence

Compare

OUTPUTSequence

OUTPUTSequence


Attack point

Attack Point

  • Find a better way of evaluation

    • Better search guidance

  • Find new encoding

    • Reduce encoding redundancy

  • Find a way to reduce destructive effect of crossover

    • Short defining length encoding

    • New crossover operator


Attack point evaluation

Attack Point : Evaluation

0/B

  • Evaluation by checking output can mislead the search process

B

A

0/A

1/A

1/B

Target Machine

0/A

B

A

0/B

1/B

1/A

Hypothesis Machine


Attack point encoding

Attack Point : Encoding

0

  • Some machines are behavioral equivalence while they differ in encoding

0

B

A

1

1

B C B C A B

C

1

0

Machine A

0

C

0

C B C B A C

A

1

1

B

1

0

Machine B


Attack point crossover

Attack Point : Crossover

  • Crossover that

    • Reduce disruption effect

      • Knowledge of linkage

      • Compact representation

    • Better understanding of underlying structure

A

A

A

A

A

A

A


Work plan

Work Plan

  • Study the works in the related fields

  • Set up a reference method

  • Develop a new method

  • Set up an experiment

    • compare new method with reference method

  • Validate and summarize the result from the experiment

  • Conclude the research

  • Write a thesis


Objective

Objective

  • To develop a better genetic algorithm method for the problem


Scope of the research

Scope of the research

  • Compare the new method with reference genetic algorithm method

  • The new method must be better than the reference method

  • The solutions from the new method must be shown to be consistency


Benefit

Benefit

  • Having a better genetic algorithm method for the problem


  • Login