EXPLORATIONS OF A CONNECTIONIST CATEGORY LEARNING MODEL

连接主义范畴学习模型的探索

• 批准号: 2392950
• 负责人: JOHN K KRUSCHKE
• 金额: $ 6.75万
• 依托单位: TRUSTEES OF INDIANA UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 1994
• 资助国家: 美国
• 起止时间: 1994-04-01 至 1999-03-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/2392950
• 关键词: association learning; attention; behavioral /social science research tag; human subject; learning; personal computers; psychological models

The phrase "category learning," as used in this proposal, refers to any situation in which a person must learn to classify stimuli, such as learning to classify lists of symptoms into the correct disease category. Theories of category learning have hypothesized various underlying representations ranging from rules to prototypes to exemplars, and various learning processes ranging from gradual association building to all-or- none rule acquisition. The proposed research explores applications of successively greater extensions of the ALCOVE model of category learning (Kruschke 1990, 1992). The ALCOVE model is a synthesis of three traditions in theories of category learning: It combines (i) exemplar-based representation with (ii) dimensional attention and (iii) error-driven learning. It is a theoretical synthesis that also quantitatively fits human-learning data in a variety of situations. The intermediate-range goal of this research is to explore the scope of applicability of ALCOVE and its extensions, thereby determining the scope of its underlying explanatory principles. The long-range goal of this research is to establish empirical evidence and theoretical accounts for relations between exemplar, rule-based, and other models of category learning. The significance of the research will be its role in covering a broad range of empirical data under a single explanatory umbrella, and in unifying alternative mechanisms and determining situations in which each seems to be dominant in human learning. All the proposed research is relevant to our knowledge of mental health insofar as it will help identify underlying learning mechanisms in normal adults. There is additional potential of direct relevance to mental health if it is found that exemplar and rule-based systems are dissociable components of category learning. For example, Pinker (1991) has argued that there are separate exemplar and rule systems in language learning, and has adduced behavioral effects of brain lesions to argue his case. The studies proposed here might lead to analogous decompositions for general category learning, which could in turn eventually have broad implications for teaching methods and therapies for both normals and the mentally challenged or brain damaged.

本提案中使用的“类别学习”一词是指任何 一个人必须学会对刺激进行分类的情况，例如 学习将一系列症状归类为正确的疾病类别。 类别学习理论已经假设了各种潜在的 从规则到原型到范例，以及各种 学习过程，从逐步建立联系到所有或- 无规则获取。 拟议的研究探讨了应用程序的连续更大的 类别学习的ALCOVE模型的扩展（Kruschke 1990，1992）。 ALCOVE模型是三种传统理论的综合， 类别学习：它结合了（i）基于范例的表示与 (ii)维度注意和（iii）错误驱动学习。这是一个 理论合成，也定量拟合人类学习数据， 各种各样的情况。 本研究的中期目标是探索 ALCOVE及其扩展的适用性，从而确定范围 它的基本解释原则。这个项目的长期目标是 研究是建立经验证据和理论解释， 范例、基于规则和其他范畴模型之间的关系 学习这项研究的意义将在于它在涵盖一个 广泛的经验数据下一个单一的解释伞，并在 统一替代机制，并确定 似乎在人类学习中占主导地位。 所有提议的研究都与我们对精神健康的认识有关 只要它将有助于确定潜在的学习机制， 成年人了还有其他可能与心理健康直接相关 如果发现范例和基于规则系统是不可分离的 类别学习的组成部分。例如，Pinker（1991）认为， 在语言学习中有独立的范例和规则系统， 并援引了大脑损伤的行为影响来为他的案件辩护。的 这里提出的研究可能会导致类似的分解一般 类别学习，这反过来又会产生广泛的影响， 为正常人和精神病患者教授方法和疗法 受到挑战或者大脑受损