Fast Information Distribution for Massively Parallel IDA* Search
نویسنده
چکیده
Search permeates all aspects of artificial intelligence including problem solving, robot motion planning, concept learning, theorem proving, and natural language understanding [12]. Because search routines are frequently also a computational bottleneck, numerous methods have been explored to increase the efficiency of search by making use of background knowledge, search macro operators, and parallel hardware for search. We introduce an algorithm for massively-parallel heuristic search, named MIDA* (Massively-parallel IncrementalDeepening A* search). In this paper we demonstrate that MIDA* offers a significant improvement in efficiency of search over serial search algorithms and MIMD parallel algorithms that can only make use of a few processors. At the heart of the MIDA* approach lies a very fast information distribution algorithm. Given information about favorable operator orderings, MIDA* can improve upon SIMD algorithms that rely on traditional information distribution techniques. In the next section, we describe the IDA* algorithm and discuss the merits of operator ordering within this paradigm. We then introduce the Fifteen Puzzle problem and the robot arm path planning problem that provide the test domains for MIDA*. Next, we describe the MIDA* algorithm and present experimental results for the Fifteen Puzzle and robot motion planning problems. Finally, we analyze the benefits of MIDA* search and discuss directions for improvement and future application.
منابع مشابه
Block-Parallel IDA* for GPUs
We investigate GPU-based parallelization of IterativeDeepening A* (IDA*). We show that straightforward threadbased parallelization techniques which were previously proposed for massively parallel SIMD processors perform poorly due to warp divergence and load imbalance. We propose Block-Parallel IDA* (BPIDA*), which assigns the search of a subtree to a block (a group of threads with access to fa...
متن کاملBlock-Parallel IDA* for GPUs (Extended Manuscript)
We investigate GPU-based parallelization of IterativeDeepening A* (IDA*). We show that straightforward thread-based parallelization techniques which were previously proposed for massively parallel SIMD processors perform poorly due to warp divergence and load imbalance. We propose Block-Parallel IDA* (BPIDA*), which assigns the search of a subtree to a block (a group of threads with access to f...
متن کاملA Modified Discreet Particle Swarm Optimization for a Multi-level Emergency Supplies Distribution Network
Currently, the research of emergency supplies distribution and decision models mostly focus on deterministic models and exact algorithm. A few of studies have been done on the multi-level distribution network and matheuristic algorithm. In this paper, random processes theory is adopted to establish emergency supplies distribution and decision model for multi-level network. By analyzing the char...
متن کاملResearch on programming languages for massively parallel processing
a research group involved in the research project \New Information Processing System on Massively Parallel Processing Principle" in the Grant-in-Aid for Scienti c Research on Priority Areas of the Ministry of Education, Science and Culture, Japan. The objective of the research is the following two points according to the top level research objective of our Massively Parallel Processing Principl...
متن کاملIDA*_MCSP: a fast exact MCSP algorithm
QoS routing has been shown to be NP-hard. A recent study of its hardness suggests that the “worst-case” may not occur in practice and thus there may exist a fast exact algorithm. In this paper, we deploy the idea of iterative deepening search and look ahead to design an exact algorithm for finding the shortest path subject to multiple constraints (the MCSP problem). The accuracy of look-ahead i...
متن کامل