Constructive Neural Networks

Constructive Neural Networks

This book presents a collection of invited works that consider constructive methods for neural networks, taken primarily from papers presented at a special th session held during the 18 International Conference on Artificial Neural Networks ...

Author: Leonardo Franco

Publisher: Springer

ISBN: 9783642045127

Category: Mathematics

Page: 293

View: 771

This book presents a collection of invited works that consider constructive methods for neural networks, taken primarily from papers presented at a special th session held during the 18 International Conference on Artificial Neural Networks (ICANN 2008) in September 2008 in Prague, Czech Republic. The book is devoted to constructive neural networks and other incremental learning algorithms that constitute an alternative to the standard method of finding a correct neural architecture by trial-and-error. These algorithms provide an incremental way of building neural networks with reduced topologies for classification problems. Furthermore, these techniques produce not only the multilayer topologies but the value of the connecting synaptic weights that are determined automatically by the constructing algorithm, avoiding the risk of becoming trapped in local minima as might occur when using gradient descent algorithms such as the popular back-propagation. In most cases the convergence of the constructing algorithms is guaranteed by the method used. Constructive methods for building neural networks can potentially create more compact and robust models which are easily implemented in hardware and used for embedded systems. Thus a growing amount of current research in neural networks is oriented towards this important topic. The purpose of this book is to gather together some of the leading investigators and research groups in this growing area, and to provide an overview of the most recent advances in the techniques being developed for constructive neural networks and their applications.
Categories: Mathematics

Constructive Neural Networks

Constructive Neural Networks

Author: Nicholas K. Treadgold

Publisher:

ISBN: OCLC:223138619

Category: Neural networks (Computer science)

Page: 586

View: 216

Categories: Neural networks (Computer science)

Executive Functions and Constructive Neural Networks

Executive Functions and Constructive Neural Networks

The present work explores how executive functions can be implemented in neural networks.

Author: John Larry Stricker

Publisher:

ISBN: UCSD:31822009435504

Category: Human information processing

Page: 228

View: 389

The present work explores how executive functions can be implemented in neural networks. Computational models such as neural networks allow researchers to develop more sophisticated conceptualizations of how executive functioning could be implemented in the brain. However, most computational models are designed only to solve a single problem rather than to solve multiple problems and integrate new and old knowledge. The problem domain for the models of the present work consists of Boolean logic expressions. These expressions easily lend themselves to implementation in neural networks while at the same time they can represent a range of problems that relate to executive functions, such as learning complementary vs. unrelated information. Network architectures and training regimes are developed that allow neural networks to solve multiple problems constructively while minimizing the impact of interference. The networks illustrate that the constructive learning of multiple problems does not require an executive controller, separate memory systems, or the constructive addition of learning resources.
Categories: Human information processing

Constructive Neural Networks

Constructive Neural Networks

This chapter presents and discusses several well-known constructive neural network algorithms suitable for constructing feedforward architectures aiming at ...

Author: Leonardo Franco

Publisher: Springer Science & Business Media

ISBN: 9783642045110

Category: Mathematics

Page: 293

View: 712

This book presents a collection of invited works that consider constructive methods for neural networks, taken primarily from papers presented at a special th session held during the 18 International Conference on Artificial Neural Networks (ICANN 2008) in September 2008 in Prague, Czech Republic. The book is devoted to constructive neural networks and other incremental learning algorithms that constitute an alternative to the standard method of finding a correct neural architecture by trial-and-error. These algorithms provide an incremental way of building neural networks with reduced topologies for classification problems. Furthermore, these techniques produce not only the multilayer topologies but the value of the connecting synaptic weights that are determined automatically by the constructing algorithm, avoiding the risk of becoming trapped in local minima as might occur when using gradient descent algorithms such as the popular back-propagation. In most cases the convergence of the constructing algorithms is guaranteed by the method used. Constructive methods for building neural networks can potentially create more compact and robust models which are easily implemented in hardware and used for embedded systems. Thus a growing amount of current research in neural networks is oriented towards this important topic. The purpose of this book is to gather together some of the leading investigators and research groups in this growing area, and to provide an overview of the most recent advances in the techniques being developed for constructive neural networks and their applications.
Categories: Mathematics

Behavior Learning with Constructive Neural Networks in Mobile Robotics

Behavior Learning with Constructive Neural Networks in Mobile Robotics

This book presents a learning system for acquiring robot behaviors by mapping sensor information directly to motor actions.

Author: Jun Li

Publisher: LAP Lambert Academic Publishing

ISBN: 3838380061

Category:

Page: 156

View: 241

In behavior-based robotics, a robot achieves a required task by using various behaviors as the building blocks for that overall task. A robot behavior in turn is a sequence of sensory states and their corresponding motor actions, and extends in time and space. Making a robot able to learn (or develop) meaningful and purposeful behaviors from its own experiences has played one of the most important roles in intelligent robotics, and have been called the hallmark of intelligence. This book presents a learning system for acquiring robot behaviors by mapping sensor information directly to motor actions. It addresses the integration of three learning paradigms, namely unsupervised learning, supervised learning, and reinforcement learning. The approach is characterized by the use of constructive artificial neural networks, Several novel techniques for robot learning using constructive radial basis function networks are introduced. The learning system is verified by a number of experiments involving a real robot learning different behaviors. It is shown that the learning system is useful as a generic learning component for acquiring diverse behaviors in mobile robots.
Categories:

Learning in Fractured Problems with Constructive Neural Network Algorithms

Learning in Fractured Problems with Constructive Neural Network Algorithms

Evolution of neural networks, or neuroevolution, has been a successful approach to many low-level control problems such as pole balancing, vehicle control, and collision warning.

Author: Nate F. Kohl

Publisher:

ISBN: OCLC:708263792

Category:

Page: 328

View: 977

Evolution of neural networks, or neuroevolution, has been a successful approach to many low-level control problems such as pole balancing, vehicle control, and collision warning. However, certain types of problems -- such as those involving strategic decision-making -- have remained difficult to solve. This dissertation proposes the hypothesis that such problems are difficult because they are fractured: The correct action varies discontinuously as the agent moves from state to state. To evaluate this hypothesis, a method for measuring fracture using the concept of function variation of optimal policies is proposed. This metric is used to evaluate a popular neuroevolution algorithm, NEAT, empirically on a set of fractured problems. The results show that (1) NEAT does not usually perform well on such problems, and (2) the reason is that NEAT does not usually generate local decision regions, which would be useful in constructing a fractured decision boundary. To address this issue, two neuroevolution algorithms that model local decision regions are proposed: RBF-NEAT, which biases structural search by adding basis-function nodes, and Cascade-NEAT, which constrains structural search by constructing cascaded topologies. These algorithms are compared to NEAT on a set of fractured problems, demonstrating that this approach can improve performance significantly. A meta-level algorithm, SNAP-NEAT, is then developed to combine the strengths of NEAT, RBF-NEAT, and Cascade-NEAT. An evaluation in a set of benchmark problems shows that it is possible to achieve good performance even when it is not known a priori whether a problem is fractured or not. A final empirical comparison of these methods demonstrates that they can scale up to real-world tasks like keepaway and half-field soccer. These results shed new light on why constructive neuroevolution algorithms have difficulty in certain domains and illustrate how bias and constraint can be used to improve performance. Thus, this dissertation shows how neuroevolution can be scaled up from learning low-level control to learning strategic decision-making problems.
Categories: