Abstract: The knowledge acquisition of mankind usually takes advantage of the known knowledge to infer relevant unknown knowledge.Machine learning can resort to mankind’s process of knowledge acquisition.From the reverse process of learning，knowledge can be viewed as the start point of the learning processing to study and apply the results of cognitive learning.The known knowledge regarded as referential knowledge produces new knowledge through equivalent and approximate equivalent transformation.Approximate equivalent transformation includes expanding process and reduction process.Equivalent transformation processes knowledge of the same hierarchies.Expanding process of approximate equivalence produces detailed-level knowledge，and reduction process produces abstract-level knowledge.Instances analysis show that expanding and reduction algorithms have high accuracy for inferring new knowledge and can produce detailed and abstract level knowledge.

Key words: cognition, machine learning, equivalence, approximate equivalence, reduction

摘要： 人类的知识获取过程通常是利用已有知识来推知相关知识的过程，机器学习可以借助这一过程，从学习的逆过程出发，把知识作为学习的出发点，来对认知学习进行研究和应用。以已有的知识作为参照知识，经过等价变换或约等价变换，产生新的知识。约等价变换过程包括知识的约等价扩充和约等价约简。等价变换过程处理同一层次的知识，约等价扩充过程产生细节级知识，约等价约简过程产生抽象级知识。实例分析表明，采用知识的约等价扩充和约简算法所产生的新知识的准确率较高，能够产生当前已有知识的细节级和抽象级新知识，为认知机器学习的知识处理提供了一定参考。

关键词: 认知, 机器学习, 等价, 约等价, 约简