RIA: Efficient and Accurate Prediction Systems for Large Scale Problems

RIA：针对大规模问题的高效、准确的预测系统

• 批准号: 9308523
• 负责人: Zoran Obradovic
• 金额: $ 16.12万
• 依托单位: Washington State University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1993
• 资助国家: 美国
• 起止时间: 1993-07-01 至 1997-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9308523&HistoricalAwards=false
• 关键词: RIA; Efficient; Accurate; Prediction; Systems

Neural network technology has the potential to solve an impressive variety of today's technological problems which conventional computers are quite powerless to deal with. The first applications of neural networks technology have been limited to relatively small problems, and therefore systems scaling was not a primary issue. Neural networks for real-life problems usually consist of a large numbers of interconnected processors working in parallel. The objective of this proposal is a deeper investigation into the computational and learning abilities of neural networks for such large scale problems. In particular, the computational models and topologies best suited for large scale problems, and efficient and accurate learning techniques for such resource- bounded parallel machine will be investigated. The results will be evaluated on two challenging real-life problems from electrical engineering and molecular biology (mixed-signal integrated circuits testing and protein structure prediction). The proposed work is based on recent results: (1) a discrete multivalued neural networks model that can be used as a tool for reasoning about the computational scaling abilities of various neural network architectures; (2) efficient neural network learning algorithms that generalize well starting with pre-existing symbolic knowledge and incorporating additional constructive learning from examples.//

神经网络技术有可能解决一个 当今各种令人印象深刻的技术问题 传统的计算机是无能为力的。 第一 神经网络技术的应用已经被限制在 相对较小的问题，因此系统扩展不是一个 首要问题。 神经网络解决现实问题通常 由大量相互连接的处理器组成， 并联 这项建议的目的是进行更深入的调查 神经网络的计算和学习能力 对于如此大规模的问题。 特别是，计算 最适合大规模问题的模型和拓扑，以及 有效和准确的学习技术，这样的资源- 有界并行机将被研究。 结果将 评估了两个具有挑战性的现实生活中的问题，从电气 工程和分子生物学（混合信号集成电路 测试和蛋白质结构预测）。 拟议的工作是 基于最近的结果：（1）离散多值神经网络 网络模型，可用作推理的工具， 各种神经网络的计算缩放能力 （2）高效的神经网络学习算法 从预先存在的符号知识开始， 并从示例中融入额外的建设性学习。//