SGER: Machine Learning by Combining Knowledge and Adaptation

SGER：结合知识和适应的机器学习

• 批准号: 9707333
• 负责人: Li Fu
• 金额: $ 1.8万
• 依托单位: University of Florida
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1997
• 资助国家: 美国
• 起止时间: 1997-03-01 至 1998-02-28
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9707333&HistoricalAwards=false
• 关键词: SGER; Machine; Learning; Combining; Knowledge; SGER; Machine; Learning; Combining; Knowledge

In this exploratory research, an incremental-learning rule- based neural network is developed using bounded weight modification and structural adaptation. Data information stored as clusters can be incrementally updated by adding cluster-based processing units to the neural network. Learning theory suggests that rule-based neural networks employing the CF-based activation function require smaller sample sizes for correct generalization than standard multilayer perceptrons. Experiments for evaluating the learning capacity and sample complexity of such systems are to be conducted. This research is intended for building expert systems with the capabilities of automatic inference and life-long learning in a wide range of domains.

在这项探索性研究中，使用有界的重量修改和结构适应开发了基于学习规则的神经网络。通过将基于群集的处理单元添加到神经网络中，可以逐步更新存储为群集的数据信息。学习理论表明，采用基于CF的激活功能的基于规则的神经网络需要比标准多层感知的更小的样本大小才能正确概括。要评估此类系统的学习能力和样本复杂性的实验。 这项研究旨在建立具有自动推理和终身学习能力的专家系统。