Seiichi Nishihara

Constructive generation of very hard 3-colorability instances

Graph colorability (COL), is a typical constraint satisfaction problem to which phase transition phenomena (PTs), are important in the computational complexity of combinatorial search algorithms. PTs are significant and subtle because, in the PT region, extraordinarily hard problem instances are found, which may require exponential-order computational time to solve. To clarify PT mechanism, many studies have been undertaken to produce very hard instances, many of which were based on generate-and-test approaches. We propose a rather systematic or constructive algorithm that repeats the embedding of 4-critical graphs to arbitrarily generate large extraordinarily hard 3-colorability instances. We demonstrated experimentally that the computational cost to solve our generated instances is of an exponential order of the number of vertices by using a few actual coloring algorithms and constraint satisfaction algorithms.

Urban Traffic Signal Control Based on Distributed Constraint Satisfaction

Urban traffic problems including traffic accidents and traffic congestion or jams have been very serious for us. Urban traffic flow simulation has been important for making new control strategies that can reduce traffic jams. In this paper, we propose a method that can dynamically control traffic signals by equivalently representing as the constraint satisfaction problem, CSP. To solve local congestion in each intersection, we define the whole system as multi- agent systems where the represented CSP is extended to distributed CSP, DCSP, in each of which variable is distributed among each intersection agent. Each intersection agent determines some signal parameters by solving the DCSP. The proposed method is implemented on our separately developed agent-oriented urban traffic simulator and applied to some roadnetworks, whose experimental simulations demonstrated that our method can effectively reduce traffic jams even in the roadnetworks where traffic jams are liable to occur.

Solving constraint-satisfaction problems by a genetic algorithm adopting viral infection

Abstract Several approximate algorithms have been reported to solve large constraint-satisfaction problems (CSPs) within a practical time. While those papers discuss techniques to escape from local optima, this paper describes a method that actively performs global searches. The present method improves the rate of search of genetic algorithms by using viral infection instead of mutation. Partial solutions of a CSP are considered to be viruses, and a population of viruses is created, as well as a population of candidate solutions. The search for a solution is conducted by crossover and infection. Infection substitutes the gene of a virus for the locus decided by the virus. Experimental results using randomly generated CSPs prove that the proposed method is faster that usual genetic algorithms at finding a solution when the constraint density of a CSP is low.

Improving performance of GP by adaptive terminal selection

Genetic Programming (GP) is an evolutionary search algorithm which searches a computer program capable of producing the desired solution for a given problem. For the purpose, it is necessary that GP system has access to a set of features that are at least a superset of the features necessary to solve the problem. However, when the feature set given to GP is redundant, GP suffers substantial loss of its efficiency. This paper presents a new approach in GP to acquire relevant terminals from a redundant set of terminals. We propose the adaptive mutation based on terminal weighting mechanism for eliminating irrelevant terminals from the redundant terminal set. We show empirically that the proposed method is effective for finding relevant terminals and improving performance of GP in the experiments on symbolic regression problems.

Multiresolution wavelet analysis of shape orientation for 3d shape retrieval

In the present paper, we propose a novel 3D shape descriptor by performing multiresolution wavelet analysis on shape orientation. We consider the spatial orientation of the polygon surfaces of a shape as important information and characterize this information by setting view planes. We then analyze these view planes by multiresolution wavelet analysis, a powerful tool used in signal processing, and lower the high resolution to low frequency domains because the high resolution contains too much information, which must be reduced in order to capture the main components. We compare the proposed descriptor to two of the best-performing descriptors on the Princeton Shape Benchmark, Spherical Harmonics Descriptor and Light Field Descriptor, and analyze the performance of the proposed descriptor from several aspects. We also compare the proposed descriptor to the Spherical Wavelet Descriptor, which won the best paper award at SMI06, a near method to our descriptor. The proposed descriptor improves the retrieval performance.

Genetic algorithms for constraint satisfaction problems

Several approximate algorithms using hill-climbing techniques and neural networks have been proposed to solve large constraint satisfaction problems (CSPs) in a practical time. In these proposals, many methods of escaping from local optima are discussed, however, there are very few methods actively perform global search. In this paper we propose a hybrid search method that combines the genetic algorithm with the min-conflicts hill-climbing (MCHC). In our method, the individual that has the fewest conflicts in the population is used as the initial value of MCHC to search locally. The detailed experimental simulation is also performed to prove that the proposed method generally gives better efficiency than the random restarting MCHC when CSPs are sparsely-connected.

An ALife approach to modeling virtual cities

This paper proposes a novel method that enables automatic modeling of virtual cities. The method makes use of L-systems to generate road networks and the genetic algorithm (GA) to generate building layouts. The road networks are composed of two types of roads-linear flow systems which are generated by using the Tree L-system and cellular networks which are generated by using the Map L-system. A generation procedure for road networks and building layouts is described. Some experimental results of generated road networks and virtual cities are also shown.

Solving constraint satisfaction problems by a genetic algorithm adopting viral infection

Several approximate algorithms have been reported to solve large constraint satisfaction problems (CSPs) in a practical time. While these papers discuss techniques to escape from local optima, the present paper describes a method that actively performs global search. The present method is to improve the rate of search of genetic algorithms using viral infection instead of mutation. The partial solutions of a CSP are considered to be viruses and a population of viruses is created as well as a population of candidate solutions. Search for a solution is conducted by crossover infection substitutes the gene of a virus for the locus decided by the virus. Experimental results using randomly generated CSPs prove that the proposed method is faster than a usual genetic algorithm in finding a solution when the constraint density of a CSP is low.

Understanding three-view drawings of mechanical parts with curved shapes

A system that interprets a given three-view engineering drawing and reconstructs 3D solid models with curved shapes is described. The solid object is confined to the 3D sheet metal object composed of planes and cutting faces consisting of lines and arcs. By giving the high priority to the cutting faces of the 3D sheet metal object in the operation of the combinatorial search of faces, one can reduce the reconstruction processing time. It is shown that 3D scenes can be successfully reconstructed by the proposed algorithm, and the efficiency of the understanding system is also proved by experiments.<<ETX>>

Interpreting engineering drawings of polyhedrons

A system is discussed that restores solid models, or a set of polyhedra consistent with a given set of three orthographic views which are input by using a facsimile device. Emphasis is on the description of two novel techniques: the line-segment extraction method and the polyhedron restoration method. The line-segment extraction is performed by using a probe, a small fragment of line, which moves along the central axis of each line segment represented by a binary image. Therefore, the method does not require any preprocessing such as noise elimination, or thinning of the given binary image. The polyhedron restoration method uses a face-oriented approach, rather than the conventional wire-frame-oriented approach. The method is basically a combinatorial search procedure that finds all of the legal combinations of possible faces.<<ETX>>