1. An Adaptive Learning Solution to the Keyboard Optimization Problem Reporter: En-ping Su 11.29 2005

2. Outline 1. Introduction 2. Pervious solutions 3. Automata solutions to Opt-Keyboard 4. Experimental results 5. Conclusion

3. 1. Introduction Ambiguous keyboards have been popularly used in touch-tone devices ex: PDA, Cellular phone, Keyboard for disability … etc. 12 abc 3 def 4 ghi 5 jkl 6 mno 7 pqrs 8 tuv 9 wxyz

4. 1. Introduction When we want to display the string “ box ”, we could just as easily have transmitted the digit string “ 269 ” But the string “ 269 ” could represent both words “ box ” and “ boy ” 12 abc 3 def 4 ghi 5 jkl 6 mno 7 pqrs 8 tuv 9 wxyz

5. 1. Introduction Above-mentioned, it could cause the ambiguous situation In this paper we consider the problem of assigning the letters of the alphabet to the various digits so as to minimize the ambiguities

6. 1.1 Problem Statement and Notation A: a finite alphabet h: the finite dictionary Ci: a set of characters list C1={} ; C2={abc} ; C3={def} C4={ghi} ; C5={jkl} ; C6={mno} C7={pqrs} ; C8={tuv} ; C9={wxyz} (h, Π) Δ(h, Π)(h, Π): the total ambiguity caused by a particular keyboard assignment(partition) Π

7. 2. Previous solutions The solution to the keyboard optimization problem is an evolutionary method based on a technique described by De Jong(1980) It ’ s an adaptive algorithm based on genetic reproduction

8. 2. Previous solutions This adaptive system is formally represented by De Jong as follows. It will be represented by a set of dictionaries It represents a set of all possible keyboards It ’ s equal to the function (h, Π) Δ(h, Π)(h, Π) Reproduction involving the crossover and mutation operations

9. 3. Automata solution to Opt- Keyboard Our solution to the problem involves the use of stochastic learning automata and such automata have various applications such as parameter optimization, statistical decision making and telephone routing

10. 3. Automata solution to Opt- Keyboard

11. Ex: A be the English alphabet and the dictionary h consist of eight word: h={ace,ale,ago,each,eels,ape,big,dig} Π(n)=((aeo),(cgl),(bd),(fgr),(km),(inpz),(tuy), (hsv),(jwx)) the mapping of the various words in h is the set {121,121,121,1128,1128,161,362,362}

12. 3. Automata solution to Opt- Keyboard 1 ={121,1128,161,362} thus (h, Π) Δ(h, Π)(h, Π) has the value 4 On processing the word X= “ ace ” it has two collisions and the set of colliding character are {c, e} Π′=((aeo),(gl),(bd),(fgr),(km),(cinpz),(t uy),(hsv),(jwx))

13. 3. Automata solution to Opt- Keyboard 1 ={161,121,121,1168,1128,161,364,362} the (h, Π) Δ(h, Π)(h, Π) has the value 2 Π(n+1) is assigned the value Π With the restriction that the number of characters per key had to be between 2 and 4 inclusive Π(n+1)=((aeo),(glp),(bd),(fgr),(km),(cinz),(t uy),(hsv),(jwx)) ′ ′

14. 4. Experimental results

17. 5. Conclusion

18. In this paper, they applied a new learning automaton solution to the keyboard problem and experimental results which were demonstrated the power of the automaton