China 
Africa 
Explore chapters and articles related to this topic
Story of Robots
Published in Junichi Takeno, Self-Aware Robots, 2022
Early famous studies on planning problems include GPS (General Problem Solver) by Newell and Simon (1963) and STRIPS (Stanford Research Institute Problem Solver) by Nilsson et al. (1971). Using GPS, given the start state S and goal state G, the procedure from S to G is automatically generated. The monkey and banana problem is a famous toy problem in the study of artificial intelligence. Like GPS, STRIPS calculates the procedure from S to G automatically based on hierarchical planning.
The Statistical Filter Approach to Constrained Optimization
Published in Technometrics, 2020
Tony Pourmohamad, Herbert K. H. Lee
We demonstrate the effectiveness of our statistical filter method on a synthetic test problem, a welded beam problem from the literature, as well as the real world pump-and-treat hydrology problem of Section 5.3. As a benchmark, we choose to compare the results of our method to the results of the methods used in Hedar (2004) and Gramacy et al. (2016). The approach used in Hedar (2004) was very different from the statistical approach we take and so we focus less on replicating the exact conditions of their experiment, and concentrate more on arriving at the same solution but in fewer objective function evaluations. However, to make a fair comparison of our methodology with Gramacy et al. (2016) we try to follow and mimic their initial conditions as closely as possible. For the toy problem, we initialize the PLMGP model with 10 random input–output pairs from before starting our optimization algorithm. Likewise, we use our surrogate model to predict outputs based on a random set of 1000 candidate locations in . Finer details for the initialization steps of the real-world hydrology computer experiment were not present in Gramacy et al. (2016) and so we chose to initialize the PLMGP model with 60 random input–output pairs from . Of note, our statistical filter method was designed to treat both the objective and constraint functions as outputs from a black box computer models, however, in Gramacy et al. (2016) they treated the objective function as being known. Therefore, to further facilitate a fair comparison of methods, we chose to also treat the objective functions as known and to only treat the constraint functions as black box outputs. This decision does not interfere with our statistical filter methodology as we may still treat the objective function as a black box output during the modeling and prediction stages, but then we shall deterministically predict the value of the objective function given that we now know the true form of it but still use our PLMGP model to predict the values of the constraint functions. In the welded beam problem (Hedar 2004), we treat both the objective and constraint functions as unknown functions, that is, arising from a black box computer model.