The paper provides a unified view of computationalintelligence in the context of software engineering. Technologies such as fuzzy sets, neural and evolutionary computing useful in software development are considered....
详细信息
The paper provides a unified view of computationalintelligence in the context of software engineering. Technologies such as fuzzy sets, neural and evolutionary computing useful in software development are considered. The links between software engineering and computationalintelligence are identified. An illustration is given in terms of a fuzzy software quality model.
The paper presents a model for hierarchical fuzzy controllers, which is a form of fuzzy gain scheduling. The underlying strategy in designing hierarchical fuzzy control systems is a selection of a global policy (globa...
详细信息
The paper presents a model for hierarchical fuzzy controllers, which is a form of fuzzy gain scheduling. The underlying strategy in designing hierarchical fuzzy control systems is a selection of a global policy (global nonlinear control) based on a collection of local controllers that are solely developed with respect to "locally" linearized models of the system. A sample hierarchical fuzzy attitude control system for a specific class of small satellites is given.
Reinforcement learning is an integral part of intelligent agent research. The development of this field, however, has been largely independent of the latest developments in neural networks. As a result, the most popul...
详细信息
Reinforcement learning is an integral part of intelligent agent research. The development of this field, however, has been largely independent of the latest developments in neural networks. As a result, the most popular designs for intelligent agents utilize neural network architectures from several years ago. This article recommends newer, proven designs for reinforcement learning. The recommended designs share historical roots with the most popular architectures in place today, allowing improved performance without radical redesign of existing agents.
Reinforcement learning is an integral part of intelligent agent research. The development of this field, however, has been largely independent of the latest developments in neural networks. As a result, the most popul...
详细信息
Reinforcement learning is an integral part of intelligent agent research. The development of this field, however, has been largely independent of the latest developments in neural networks. As a result, the most popular designs for intelligent agents utilize neural network architectures from several years ago. The article recommends newer, proven designs for reinforcement learning. The recommended designs share historical roots with the most popular architectures in place today, allowing improved performance without radical redesign of existing agents.
The paper discusses convergence issues when training adaptive critic designs (ACD) to control dynamic systems expressed as Markov sequences. We critically review two published convergence results of critic based train...
详细信息
The paper discusses convergence issues when training adaptive critic designs (ACD) to control dynamic systems expressed as Markov sequences. We critically review two published convergence results of critic based training and propose to shift emphasis towards more practically valuable convergence proofs. We show a possible way to prove convergence of ACD training.
One of the problems of existing fuzzy-neural approaches is that the logic nature of the structure is often lost, i.e., what is being processed by the neural networks becomes irrelevant. To retain this logic content wh...
详细信息
One of the problems of existing fuzzy-neural approaches is that the logic nature of the structure is often lost, i.e., what is being processed by the neural networks becomes irrelevant. To retain this logic content while benefiting from the advantage of integrating fuzzy set and neural network approaches, we propose in this paper a fuzzy neural network which supports fuzzy inference mechanisms by being based exclusively on logic implication neurons. A supervised learning method involving an equality performance measure and an online update delta rule (gradient-based) learning procedure is used. An experimental study involving Wolfer's sunspot numbers is carried out, demonstrating fast convergence accompanied by explicit format of the inference network.
We present a procedure for obtaining the derivatives used in training a recurrent network that combines in a unified framework the techniques of backpropagation through time and derivative adaptive critics. The result...
详细信息
We present a procedure for obtaining the derivatives used in training a recurrent network that combines in a unified framework the techniques of backpropagation through time and derivative adaptive critics. The resulting formulation is consistent with previous descriptions, but has the advantage of allowing the mentioned techniques to be used together in a proportion that is appropriate to a given problem.
By studying adaptive critic designs (ACD) from the standpoint of practical use in training neural networks, we expect to establish the types of problems for which ACD might be preferable to more established methods. T...
详细信息
By studying adaptive critic designs (ACD) from the standpoint of practical use in training neural networks, we expect to establish the types of problems for which ACD might be preferable to more established methods. To restrict the scope, we have chosen to concentrate on applying ACD, specifically derivative critics, to the training of recurrent networks (L.A. FeldKamp et al., 1997). This is actually less restrictive than it may appear; many problems, including controller training, can be posed as optimizing some or all of the weights of a recurrent network. An immediate benefit of this focus has been to clarify the relationship between the derivatives that result from backpropagation through time (BPTT) and the quantities that derivative critics are expected to deliver. At the same time, many questions have been raised, such as that of the critic representation that best balances accuracy against the number of time steps required for adaptation. Because our formulation permits BPTT and derivative critics to be used together or separately, we expect that experience with a variety of problems will further clarify the various tradeoffs and suggest situations in which critics may be used to particular advantage.
In a fuzzy number neural network, the inputs, weights, and outputs are general fuzzy numbers. The requirement that F~/sup /spl alpha/(1)//spl sub/F~/sup /spl alpha/(2/) whenever /spl alpha/(1)>/spl alpha/(2) impose...
详细信息
In a fuzzy number neural network, the inputs, weights, and outputs are general fuzzy numbers. The requirement that F~/sup /spl alpha/(1)//spl sub/F~/sup /spl alpha/(2/) whenever /spl alpha/(1)>/spl alpha/(2) imposes an enormous number of constraints on the weight parameterizations during training. This problem can be solved through a careful choice of weight representation. This new representation is unconstrained, so that standard neural network training techniques may be applied. Unfortunately, fuzzy number neural networks still have many parameters to pick during training, since each weight is represented by a vector. Thus moderate to large fuzzy number neural networks suffer from the usual maladies of very large neural networks. In this paper, we discuss a method for effectively reducing the dimensionality of networks during training. Each fuzzy number weight is represented by the endpoints of its /spl alpha/-cuts for some discretization 0/spl les//spl alpha//sub 1/
暂无评论