CSC 242 Lecture Notes - Lecture 6: Beam Search, Mutation
Document Summary
Local beam search (another way of improving random search) This time, we do not rely on randomness. Keep track of k states, for a constant k, instead of just 1 state. If we have k states, we can: run search on k states in parallel, generate successors (k*b, keep the best k of them k is width of beam . We need to remember that local search is complete, and is bound to have problems. In this case, if the goal state requires a few bad decisions, we will never find the answer, and we need another tool to find the answer. George"s take: it is essentially a local beam search with a new way of generating. Fyi: section 4. 2 (not much related to this unit, informative reading) As for nfas, an action can lead to several different states. This would be determined by your evaluation function, as in nfas.