- 作者: 洪宗貝;曾憲雄
- 作者服務機構: 國立交通大學資訊工程研究所; 國立交通大學資訊科學研究所
- 中文摘要: 在本篇論文中,我們主要是討論如何將平行處理的技術應用在機器學習的領域上。機器學習對於建立專家系統時的知識獲取相當重要,它可直接從所使用的例子中獲取所要的觀念。根據處理方式的不同,平行式機器學習可分成兩大類:平行式由上而下學習策略及平行式由下而上學習策略。針對每一類學習策略,我們提出了平行學習模式和 其適合的平行架構,並且舉例?明之。以平行式由上而下學習策略而言,我們利用工作分配的方式將學習工作平均分配於MIMD的計算機架構,並舉了一個相當著名的學習策略ID3來?明其做法。至於平行式由下而上學習策略,我 們利用了演算法中的各個擊破方式(Divide and Conpuer)建構於SIMD的計算機架構中,並舉了另一個相當著名的學習 策略Version Space來?明其做法。雖然我們所提的平行學習模式不保證適用於所有的學習方法,但卻對平行式機器學 習這個領域做了首次的研究探討。
- 英文摘要: Learning general concepts from a set of training instances has become increasingly important for artificial intelligence researchers on constructing knowledge based systems. In this paper, we attempt to apply the technique of parallel processing to concept learning. The learning strategies can be divided into two classes: top-down learning strategies and bottom-up learning strategies, and therefore, based upon the partition of learning tasks on multiple processors and the principle of divide-and-conquer, two corresponding parallel learning models are proposed. Further, we will show that these two models can be easily embedded into two practical and commonly used architectures, respectively: the mimd shared-memory architecture and the simd shared-memory architecture. Finally, based upon our models, the ID3 and the version space learning strategies are parallelized to show how a parallel top-down learning or a parallel bottom-up learning strategy can work well. The models we propose cannot be guaranteed to fit all top-down learning strategies and all bottom-up learning strategies, but they provide a principle and insight into aspects of Parallel Machine Learning.
- 中文關鍵字: top-down learning; bottom-up learning; parallel learning; ID3; version space; MIMD shared-memory architecture; SIMD shared memory architecture
- 英文關鍵字: --