The SIMD parallelism of the Connection Machine (eM) allows the construction of neural network simulations by the use of simple data and control structures. Two approaches are described which allow parallel computation of a model's nonlinear functions, parallel modification of a model's weights, and parallel propagation of a model's activation and error. Each approach also allows a model's interconnect structure to be physically dynamic. A Hopfield model is implemented with each approach at six sizes over the same number of CM processors to provide a performance comparison.

The Connection Machine introduces a new type of parallel computer which may lead to radically new ways of thinking about computing, particularly in Artificial Intelligence. The author, Daniel Hillis, is the designer of the machine and the founder of Thinking Machines Corporation, a company committed to building "Connection Machines." Hillis' book describes the Connection Machine and the issues surrounding its design. The Connection Machine is like an active memory made up of thousands, potentially millions, of small, simple, processors working simultaneously, each with its own tiny memory. This memory is controlled by a conventional, sequential, host computer.

Within the time of a human generation, computer technology will be capable of producing computers with as many artificial neurons as there are neurons in the human brain. Within two human generations, intelligists (AI researchers) will have discovered how to use such massive computing capacity in brainlike ways. This situation raises the likelihood that twenty-first century global politics will be dominated by the question, Who or what is to be the dominant species on this planet? This article discusses rival political and technological scenarios about the rise of the artilect (artificial intellect, ultraintelligent machine) and launches a plea that a world conference be held on the socalled "artilect debate." Many years ago, while reading my first book on molecular biology, I realized not only that living creatures, including human beings, are biochemical machines, but also that one day, humanity would sufficiently understand the principles of life to be able to reproduce life artificially (Langton 1989) and even create a creature more intelligent than we are.

Parallelizable optimization techniques are applied to the problem of learning in feedforward neural networks. In addition to having superior convergence properties, optimization techniques such as the Polak Ribiere method are also significantly more efficient than the Backpropagation algorithm. These results are based on experiments performed on small boolean learning problems and the noisy real-valued learning problem of handwritten character recognition. 1 INTRODUCTION The problem of learning in feedforward neural networks has received a great deal of attention recently because of the ability of these networks to represent seemingly complex mappings in an efficient parallel architecture. This learning problem can be characterized as an optimization problem, but it is unique in several respects. Function evaluation is very expensive. However, because the underlying network is parallel in nature, this evaluation is easily parallelizable.

Parallelizable optimization techniques are applied to the problem of learning in feedforward neural networks. In addition to having superior convergence properties, optimization techniques such as the Polak Ribiere method are also significantly more efficient than the Backpropagation algorithm. These results are based on experiments performed on small boolean learning problems and the noisy real-valued learning problem of handwritten character recognition. 1 INTRODUCTION The problem of learning in feedforward neural networks has received a great deal of attention recently because of the ability of these networks to represent seemingly complex mappings in an efficient parallel architecture. This learning problem can be characterized as an optimization problem, but it is unique in several respects. Function evaluation is very expensive. However, because the underlying network is parallel in nature, this evaluation is easily parallelizable.

Parallelizable optimization techniques are applied to the problem of learning in feedforward neural networks. In addition to having superior convergenceproperties, optimization techniques such as the Polak Ribiere method are also significantly more efficient than the Backpropagation algorithm.These results are based on experiments performed on small boolean learning problems and the noisy real-valued learning problem of handwritten character recognition. 1 INTRODUCTION The problem of learning in feedforward neural networks has received a great deal of attention recently because of the ability of these networks to represent seemingly complex mappings in an efficient parallel architecture. This learning problem can be characterized as an optimization problem, but it is unique in several respects. Function evaluation is very expensive. However, because the underlying network is parallel in nature, this evaluation is easily parallelizable. In this paper, we describe the network learning problem in a numerical framework and investigate parallel algorithms for its solution. Specifically, we compare the performance of several parallelizable optimization techniques to the standard Back-propagation algorithm. Experimental results show the clear superiority of the numerical techniques. 2 NEURAL NETWORKS A neural network is characterized by its architecture, its node functions, and its interconnection weights. In a learning problem, the first two of these are fixed, so that the weight values are the only free parameters in the system.

INTRODUCTION TO A SYSTEM FOR IMPLEMENTING NEURAL NET CONNECTIONS ON SIMD ARCHITECTURES Sherryl Tomboulian Institute for Computer Applications in Science and Engineering NASA Langley Research Center, Hampton VA 23665 ABSTRACT Neural networks have attracted much interest recently, and using parallel architectures to simulate neural networks is a natural and necessary application. The SIMD model of parallel computation is chosen, because systems of this type can be built with large numbers of processing elements. However, such systems are not naturally suited to generalized communication. A method is proposed that allows an implementation of neural network connections on massively parallel SIMD architectures. The key to this system is an algorithm that allows the formation of arbitrary connections between the "neurons". A feature is the ability to add new connections quickly. It also has error recovery ability and is robust over a variety of network topologies. Simulations of the general connection system, and its implementation on the Connection Machine, indicate that the time and space requirements are proportional to the product of the average number of connections per neuron and the diameter of the interconnection network.

Two approaches are described which allow parallel computation of a model's nonlinear functions, parallel modification of a model's weights, and parallel propagation of a model's activation and error. Each approach also allows a model's interconnect structure to be physically dynamic. A Hopfield model is implemented with each approach at six sizes over the same number of CM processors to provide a performance comparison. INTRODUCflON Simulations of neural network models on digital computers perform various computations by applying linear or nonlinear functions, defined in a program, to weighted sums of integer or real numbers retrieved and stored by array reference. The numerical values are model dependent parameters like time averaged spiking frequency (activation), synaptic efficacy (weight), the error in error back propagation models, and computational temperature in thermodynamic models. The interconnect structure of a particular model is implied by indexing relationships between arrays defined in a program. On the Connection Machine (CM), these relationships are expressed in hardware processors interconnected by a 16-dimensional hypercube communication network. Mappings are constructed to defme higher dimensional interconnectivity between processors on top of the fundamental geometry of the communication network.

INTRODUCTION TO A SYSTEM FOR IMPLEMENTING NEURAL NET CONNECTIONS ON SIMD ARCHITECTURES Sherryl Tomboulian Institute for Computer Applications in Science and Engineering NASA Langley Research Center, Hampton VA 23665 ABSTRACT Neural networks have attracted much interest recently, and using parallel architectures to simulate neural networks is a natural and necessary application. The SIMD model of parallel computation is chosen, because systems of this type can be built with large numbers of processing elements. However, such systems are not naturally suited to generalized communication. A method is proposed that allows an implementation of neural network connections on massively parallel SIMD architectures. The key to this system is an algorithm that allows the formation of arbitrary connections between the "neurons". A feature is the ability to add new connections quickly. It also has error recovery ability and is robust over a variety of network topologies. Simulations of the general connection system, and its implementation on the Connection Machine, indicate that the time and space requirements are proportional to the product of the average number of connections per neuron and the diameter of the interconnection network.

Two approaches are described which allow parallel computation of a model's nonlinear functions, parallel modification of a model's weights, and parallel propagation of a model's activation and error. Each approach also allows a model's interconnect structure to be physically dynamic. A Hopfield model is implemented with each approach at six sizes over the same number of CM processors to provide a performance comparison. INTRODUCflON Simulations of neural network models on digital computers perform various computations by applying linear or nonlinear functions, defined in a program, to weighted sums of integer or real numbers retrieved and stored by array reference. The numerical values are model dependent parameters like time averaged spiking frequency (activation), synaptic efficacy (weight), the error in error back propagation models, and computational temperature in thermodynamic models. The interconnect structure of a particular model is implied by indexing relationships between arrays defined in a program. On the Connection Machine (CM), these relationships are expressed in hardware processors interconnected by a 16-dimensional hypercube communication network. Mappings are constructed to defme higher dimensional interconnectivity between processors on top of the fundamental geometry of the communication network.