An artificial neural network (ANN) is currently used in multiple different applications such as bio-medicine, finance, Internet, and mobile networks. Since their inception, many advances have taken place introducing n...
详细信息
An artificial neural network (ANN) is currently used in multiple different applications such as bio-medicine, finance, Internet, and mobile networks. Since their inception, many advances have taken place introducing new models and features. Such progress resulted in different ANN models and most importantly different types of implementation, which vary from software (SW) to hardware (HW) following specific development principles. Researchers have been working significantly the last decade in this area tackling with different aspects of ANN's implementations. In this survey, we present the progress of ANN in terms of implementation as part of computing platforms. Thus, we present the ANN-enabled computing platforms in terms of algorithmic models, computing architectures, and SW/HW implementations. This work concludes with open challenges and lessons learned in order to summarize what is potentially useful for further research in the area of ANN computing platforms with a wide spectrum of applications. An artificial neural network (ANN) is considered the key element of future computing systems applied to different domains. While the algorithmic design of an ANN is one of the major engineering elements, the implementation of ANN is equally important with many difficulties that should be overcome by future engineers. This survey aims to provide a comprehensive tutorial about the ANN-enabled computing systems, i.e., computing architectures with embedded artificial intelligence (AI). Starting with the ANN models and their applications, the survey provides a taxonomy of the types of ANN computing systems. Both SW and HW implementations are provided for each of those types, which highlight the key architectural elements as well as the performance of the ANN-enabled computing systems. Open challenges and lessons learned follow to provide a discussion for future research in the area of AI computing systems.
In this paper we propose a novel recursive algorithm that models the neighborhood mechanism, which is commonly used in self-organizing neural networks (NNs). The neighborhood can be viewed as a map of connections betw...
详细信息
In this paper we propose a novel recursive algorithm that models the neighborhood mechanism, which is commonly used in self-organizing neural networks (NNs). The neighborhood can be viewed as a map of connections between particular neurons in the NN. Its relevance relies on a strong reduction of the number of neurons that remain inactive during the learning process. Thus it substantially reduces the quantization error that occurs during the learning process. This mechanism is usually difficult to implement, especially if the NN is realized as a specialized chip or in Field Programmable Gate Arrays (FPGAs). The main challenge in this case is how to realize a proper, collision-free, multi-path data flow of activations signals, especially if the neighborhood range is large. The proposed recursive algorithm allows for a very efficient realization of such mechanism. One of major advantages is that different learning algorithms and topologies of the NN are easily realized in one simple function. An additional feature is that the proposed solution accurately models hardwareimplementations of the neighborhood mechanism. (C) 2015 Elsevier Inc. All rights reserved.
The concept of adaptivity is of paramount importance in the design of future communication systems, in which a careful exploitation of the limited available resources (bandwidth, power, etc.) is required. The potentia...
详细信息
The concept of adaptivity is of paramount importance in the design of future communication systems, in which a careful exploitation of the limited available resources (bandwidth, power, etc.) is required. The potential of channel-adaptive transmission has been already recognized 30 years ago, but it did not receive much interest at that time. In the last decade, the advent of feasible software radio systems, and hence, the availability of fast flexible and reconfigurable transceivers has renewed interest in adaptive techniques, which include adaptive modulation and coding, adaptive antennas and adaptive equalization techniques. This paper focuses on adaptive modulation and adaptive error control mechanisms. Basic concepts are highlighted and an overview on the achieved results and new trends in this research area are presented. Some results from information theory are also presented, which show the limitations of these techniques and motivate further research on the practical and design issues that have to be addressed to enable performance to reach close to the theoretical limit. Copyright (C) 2002 John Wiley Sons, Ltd.
暂无评论