Presentation is loading. Please wait.

Presentation is loading. Please wait.

ขั้นตอนวิธีเชิงพันธุกรรมสำหรับ การอนุมานเครื่องจักรสถานะ จำกัด อาจารย์ที่ปรึกษาวิทยานิพนธ์ รศ. ดร. ประภาส จงสถิตย์วัฒนา ประธานกรรมการ ศ. ดร. ชิดชนก เหลือสินทรัพย์

Similar presentations


Presentation on theme: "ขั้นตอนวิธีเชิงพันธุกรรมสำหรับ การอนุมานเครื่องจักรสถานะ จำกัด อาจารย์ที่ปรึกษาวิทยานิพนธ์ รศ. ดร. ประภาส จงสถิตย์วัฒนา ประธานกรรมการ ศ. ดร. ชิดชนก เหลือสินทรัพย์"— Presentation transcript:

1 ขั้นตอนวิธีเชิงพันธุกรรมสำหรับ การอนุมานเครื่องจักรสถานะ จำกัด อาจารย์ที่ปรึกษาวิทยานิพนธ์ รศ. ดร. ประภาส จงสถิตย์วัฒนา ประธานกรรมการ ศ. ดร. ชิดชนก เหลือสินทรัพย์ กรรมการ ผศ. ดร. บุญเสริม กิจศิริกุล ผศ. ดร. ณชล ไชยรัตนะ เสนอโดย นายนัทที นิภานันท์ เลขประจำตัว

2 Introduction Inference of FSM Inference of FSM from observed input/output from observed input/output Mimic the target machine Mimic the target machine Target Machines INPUTOUTPUT Learning Method Hypothesis Machine ≡ ? ? ?

3 Presentation Outline Claim Claim Legal stuff Legal stuff Some details of claim Some details of claim Conviction Conviction Experiment Experiment Analysis Analysis Conclusion Conclusion Extras Extras Summary Summary

4 Hypothesis A new genetic algorithm proposed in this thesis is a better way to solve the problem of finite state machine inference than the former genetic algorithm A new genetic algorithm proposed in this thesis is a better way to solve the problem of finite state machine inference than the former genetic algorithm

5 Legal stuff: Objective To develop a better genetic algorithm for the problem To develop a better genetic algorithm for the problem

6 Legal Stuff: Scope Compare the new method with reference genetic algorithm Compare the new method with reference genetic algorithm The new method must be shown to be better than the reference method The new method must be shown to be better than the reference method The solutions from the new method must be shown to be consistent The solutions from the new method must be shown to be consistent

7 Former GA (REF) Encode δ and λ in bit string Single point crossover Evaluate by counting similar output bit INPUT Sequence OUTPUT Sequence Hypothesis Machine Hypothesis OUTPUT OUTPUT Sequence Compare Next State Output Next State Output Next State Output Next State Output 0-transition1-transition State 0 0-transition1-transition State N...

8 New GA New evaluation & encoding New evaluation & encoding NEW1 method NEW1 method New crossover operator New crossover operator NEW2 method NEW2 method

9 New Evaluation Old evaluation can mislead the search Old evaluation can mislead the search Correct Correct δ under wrong λ will result in totally wrong score A B 0/A 0/B 1/B 1/A Target Machine A B 0/B 0/A 1/A 1/B Hypothesis Machine

10 New Evaluation Main idea Main idea Each transition is evaluated by some IO pairs Each transition is evaluated by some IO pairs Why make-then-ask? Why make-then-ask? Perform local search on each output value Perform local search on each output value X 0/A IO Sequence: Evaluate particular state X by (0,B) (0,B) (0,A) X 0/?  0/B old method fixes output (A) it is only 1/3 correct new method adjusts its output according to IO. It choose B to get 2/3 correct

11 New Evaluation: Example XY (c) (a) (b) (d) Input : Output : a c d a d a d Evaluation value = = 6

12 Output Definition Output: (a) (a)  0 (b) (b)  N/A (any arbitrary value) (c) (c)  1 (d) (d)  0 XY (c) (a) (b) (d)

13 New Encoding Encoding λ is futile omitted Next State 0-transition1-transition State 0 0-transition1-transition State N...

14 New Crossover The encoding scheme introduces chance of not having a tight linkage The encoding scheme introduces chance of not having a tight linkage high destructive effect high destructive effect ABCDEFG

15 New Crossover Choose two parents, find the best one Choose two parents, find the best one Rearrange state according to DFS order Rearrange state according to DFS order discard inaccessible state discard inaccessible state Perform single point crossover on the new list of states Perform single point crossover on the new list of states

16 New Crossover: Example ABCDEFG ABCDEG

17 Experiment To compare performance of REF, NEW1 and NEW2 To compare performance of REF, NEW1 and NEW2 Measure number of generation used, time used and successful run Measure number of generation used, time used and successful run

18 FSM Setup of the experiment FSM Generator FSM Target FSM IO Sequence Generator FSM IO Seq Set REF NEW1 NEW2 Hypothesis Machine Input Algorithm Output

19 Experimental Result

20

21 Experimental Result : Summary (Generation)REFNEW1NEW2 Total749,246527,488505,780 Relative100%70.40%67.51% Best03135 Successful Runs (Time)REFNEW1NEW2Total64,35046,39351,801 Relative100%72.10%80.50% Best25226

22 Analysis New evaluation & coding = local search New evaluation & coding = local search Search space reduction Search space reduction Intron from inaccessible state Intron from inaccessible state Schema preservation Schema preservation Intron nullifying Intron nullifying Intron from inaccessible state Intron from inaccessible state intron from unexercised state intron from unexercised state

23 Search Space Reduction

24

25

26 Search Space Reduction : Summary ExperimentAA1A2 Avg. Generation Used 70.40%61.60%54.12% Avg. Time Used 72.09%55.16%49.43% Number of problem Solved %135.62%150.64%

27 Additional Experiment Comparison between NEW2 and heuristic based method, red-blue Comparison between NEW2 and heuristic based method, red-blue compare correctness of result when the size of training set is reduced compare correctness of result when the size of training set is reduced using cross validation method using cross validation method correctness = proportion of correctly identified data on test set correctness = proportion of correctly identified data on test set Training set : IO Sequence Training set : IO Sequence length: 5  35 (step 2) length: 5  35 (step 2) number: 6  36 (step 6) number: 6  36 (step 6)

28 Heuristic Method : red-blue Using heuristic in search Using heuristic in search fast fast scalable scalable No restriction on the size of the hypothesis No restriction on the size of the hypothesis

29 Correctness vs. Sample Size

30 Additional Experiment: Analysis shorter description of hypothesis is more preferable shorter description of hypothesis is more preferable Occam’s Razor Occam’s Razor

31 Size of hypothesis vs. Sample Size

32 What have been done? A genetic algorithm for finite state machine inference problem is presented A genetic algorithm for finite state machine inference problem is presented It is empirically shown that the proposed method is better than former methods It is empirically shown that the proposed method is better than former methods

33 What can be extended by others? Practical Issues Practical Issues Better linkage awareness Better linkage awareness Chromosome representation Chromosome representation Theoretical Issues Theoretical Issues Effect of intron Effect of intron Formal analysis of preference of short hypothesis Formal analysis of preference of short hypothesis

34 What would you like to ask?


Download ppt "ขั้นตอนวิธีเชิงพันธุกรรมสำหรับ การอนุมานเครื่องจักรสถานะ จำกัด อาจารย์ที่ปรึกษาวิทยานิพนธ์ รศ. ดร. ประภาส จงสถิตย์วัฒนา ประธานกรรมการ ศ. ดร. ชิดชนก เหลือสินทรัพย์"

Similar presentations


Ads by Google