Witness Algorithm Presented by Alp Sarda Witness Algorithm
Witness Algorithm Presented by Alp Sardağ
Witness Algorithm • Like other algorithms, start with a set of vectors, V*t-1={ 0(t-1), . . . , M(t-1)}. – Sayfa 81 deki ilk formül • The output of the algorithm will be a set of vectors, V*t-1={ 0(t-1), . . . , N(t-1)}.
Witness Algorithm • Let Qat the set of vectors representing PLWC value function, for performing action a at time t and performing optimally there after. • Qat = { 0 a(t), . . . , Na(t)} • There will be a separate Qat for each action a. – Sayfa 81 10 a kadar olan formüller
Witness Algorithm • The algorithm first constructs the Qat sets • Then constructs the desired V*t set of vectors. • In constructing Qat set, the algorithm incrementally build up. • Note that: – Sayfa 81 son formül
Constructing Qat • Begin with an empty set Qat, choose any belief point and construct the vector using: – Formula 10 buraya kopyalanacak. • Determine if there exist a belief state * (like Sondik’s): – Sayfa 82 formüllerin ilk iki satırı kopyala. • Once the new vector found, it is added to the list of vectors.
Constructing Qat • How the algorithm finds the new belief point? • Sayfa 83 ün sonu ve 84 ün 4 formülü kopyalanacak.
Constructing Qat • Sayfa 85 teki LP kopyalanacak.
Constructing Qat • The algorithm for constructing Qat is as follows: • Sayfa 85 pseudocode copy paste edilecek.
Witness Algorithm Reduction Phase • Once all Qat sets found for each action a, then combine them to form the set V*t that is the optimal value function. • The algorithm uses reduction phase of Monahan’s algorithm. – Monahan’s reduction LP si kopyalanacak.
- Slides: 9