Acquisition of a Mini-Super Computer for Modeling Normal and Discordered Cognition.

该提案寻求为进一步开发和评估人类认知过程若干不同方面的并行分布式处理(PDP)模型所需的工具提供资金。我们试图研究一类PDP模型，在该模型中，加工是通过大量内在可变的处理单元的相互作用进行的，这些处理单元的激活基于它们从其他单元输入的时间平均。支配熟练处理的知识储存在单元之间的连接的强度中，而作为熟练处理的基础的知识的获得是通过逐步调整由处理经验触发的连接的强度来进行的。以前的工作已经显示了这种方法在许多应用中的潜力，但要想继续取得进展，就需要增加计算资源。在某些情况下，不可能探索一系列实际多样的培训实例，从而限制了测试建立人类业绩数据模型的原则的充分性的机会。在其他情况下，已经对处理过程做出了简化的假设，现在很明显，这些简化可能导致与实验数据的不匹配。在这两种情况下，评估模型能够解释实验数据的程度的能力都受到了影响。因此，我们寻求升级我们现有的5岁的向量化‘迷你超级计算机’到更新的，更强大的模型。

This proposal seeks funds for instrumentation needed for the further development and assessment of parallel distributed processing (PDP) models of a number of different aspects of human cognitive processes. We seek to study a class of PDP models in which processing occurs through the interaction of a large number of intrinsically variable processing units whose activation is based on a time-average of their input from other units. The knowledge that governs skilled processing is stored in the strengths of the connections among the units, and the acquisition of the knowledge that underlies skilled processing occurs through gradual adjustments in the strengths of the connections triggered by processing experience. Previous work has shown the potential of the approach for a number of applications, but continued progress requires an increase in computational resources. In some cases, it has not been possible to explore a realistically diverse range of training examples, thus limiting the opportunity to test the adequacy of the principles to model human performance data. In other cases, simplified assumptions about processing have been made, and it is now clear that these simplifications can result in inadequate fits to experimental data. In both kinds of cases, the ability to assess the degree to which the models can account for experimental data has been compromised. We therefore seek an upgrade to our existing 5-year old vectorizing 'minisupercomputer' to a newer, more powerful model.