An Efficient Linear Language for Parallel Symbolic Processing

下一代智能系统和机器人将需要并行计算机的能力，然而大多数新型并行语言主要面向数值计算--例如信号处理--其中数据值是基本的数值样本。这些计算模型对于涉及人工智能应用所需的高度结构化符号数据的计算来说，失败得很惨。“线性”语言提供了一种优雅而直观的并行计算“机械”模型，在该模型中，结构化的值与基本值一样简单地传输。这些“线性”数据结构保留了传输高度结构化信息的能力，但避免了混叠和共享的问题。现有线性语言的效率为新的实现技术带来了希望。在第一阶段，我们将研究我们的新实现方法的能力，以提供非线性共享数据结构的最大效率，同时保留优雅的线性数据模型。如果成功，我们的并行线性语言的第二阶段实现将提供高效的执行，以及优雅和直观的编程模型。这一组合应该会提高利用MIMD并行计算机进行智能应用软件开发的生产率。//

Next-generation intelligent systems and robots will require the power of parallel computer, yet most novel parallel languages are geared primarily for numerical calculation--e.g., signal processing--in which the data values are elementary numerical samples. These models of computation fail miserably for calculations involving the highly-structured symbolic data required for artificial intelligence application. A "linear" language provides an elegant and intuitive "mechanical" model of parallel computation in which structured values are transmitted as simply as elementary values. These "linear" data structures retain the ability to transmit highly structured information, yet avoid the problems of aliasing and sharing. The efficiency of existing linear language promising new implementation techniques. In Phase I, we will investigate the ability of our novel implementation approach to provide most of the efficiency of nonlinear shared data structures while preserving the elegant linear data model. If successful, our Phase II implementation of a parallel linear language will provide efficient execution, together with an elegant and intuitive programming model. This combination should increase the productivity of software development for intelligent applications utilizing MIMD parallel computers.//