墙的另一面——图灵模型更深层次的思考

Another Side of the Wall——Deeper Thinking in Turing Model

最近几年,计算机体系迈向多处理器结构道路。然而,冯诺曼机器主导的多核结构,令我们处在存储墙错误的一面。地址参数引入的冗余,降低了处理器的效率,成为图灵-冯诺曼模型的致命要害。对图灵模型作更深一层思考,以信息变换统一了冯诺曼机器程序变换和神经网络变换,分析了两种变换的异同及其优劣。提出以微核为基础,并按变换的成熟程度,向灵活的可编程的冯氏机器或高速的神经网络分化。模拟生物神经系统的进化,构建为人们服务的智能机器。2010年9月15日,美国波士敦的高性能嵌入式计算(HPEC)研讨会上,耶鲁大学的欧亨尼奥.卡鲁塞伊罗教授发表了一个基于人类视觉系统的高性能计算机"神经流"(NeuFlow),其体系结构利用了与本文的仿生电脑十分相似的概念。

Computer has entered into the era of multi-processor structure.However,multicore puts us on the wrong side of the wall.The memory address parameters introduce extra-redundant,decrease system efficiency.This paper proposed a new kind of model by considering intelligent processes as a transformation on information.This model unifies the Von-Neumann machine program and neural network transformation.The paper analysed the similarities and differences of these two process and proposed a computer system based on micro-kernel architecture and diverted into Von-Neumann or neural transform according to the maturity and speed requirement.Imitating biological organism intellect,this kind of computer,serves as a tool for human beings.Professor Eugenio Culurciello of Yale,with Yann LeCun of NYU,presented a high performance Embedded computer,NeuFlow,on HPEC workshop 2010-9-15 in Boston,its system architecture concepts are quire similar to the Pseudo-Organic computer proposed in this paper.