智能系统学报
智能繫統學報
지능계통학보
CAAI TRANSACTIONS ON INTELLIGENT SYSTEMS
2014年
3期
336-342
,共7页
同调学习%同调代数%机器学习%边缘划分%边缘同调学习%邻域同调学习算法%邻域复形图%相似性
同調學習%同調代數%機器學習%邊緣劃分%邊緣同調學習%鄰域同調學習算法%鄰域複形圖%相似性
동조학습%동조대수%궤기학습%변연화분%변연동조학습%린역동조학습산법%린역복형도%상사성
homology learning%homology algebra%machine learning%margin partitioning%margin homology learn-ing%neighborhood homology learning algorithm%neighborhood complex graphs%similarity
目前已有的边缘学习算法对边缘可变的数据划分问题存在一些不足，这些算法在分类过程中不能有效地保证数据的结构特征不变。因而文章首先通过引进同调代数中的单形划分理论，从机器学习的角度对分类问题中的边缘划分进行研究，提出了一种邻域同调学习算法。算法给出了图形的邻域复形的构造方法和判断2个给定图形相似性的判定标准。最后通过在USPS_ALL手写数字集数据库和MPEG7 CE图像库上与SVM、TVQ算法的对比实验验证了本算法的有效性。
目前已有的邊緣學習算法對邊緣可變的數據劃分問題存在一些不足，這些算法在分類過程中不能有效地保證數據的結構特徵不變。因而文章首先通過引進同調代數中的單形劃分理論，從機器學習的角度對分類問題中的邊緣劃分進行研究，提齣瞭一種鄰域同調學習算法。算法給齣瞭圖形的鄰域複形的構造方法和判斷2箇給定圖形相似性的判定標準。最後通過在USPS_ALL手寫數字集數據庫和MPEG7 CE圖像庫上與SVM、TVQ算法的對比實驗驗證瞭本算法的有效性。
목전이유적변연학습산법대변연가변적수거화분문제존재일사불족，저사산법재분류과정중불능유효지보증수거적결구특정불변。인이문장수선통과인진동조대수중적단형화분이론，종궤기학습적각도대분류문제중적변연화분진행연구，제출료일충린역동조학습산법。산법급출료도형적린역복형적구조방법화판단2개급정도형상사성적판정표준。최후통과재USPS_ALL수사수자집수거고화MPEG7 CE도상고상여SVM、TVQ산법적대비실험험증료본산법적유효성。
At present, the existing margin learning algorithms still have some affects when attempting to solve the data partitioning problem of variable margins .These algorithms can not effectively maintain the structure feature of datas in classification ..At present , the existing margin learning algorithms still have defects when attempting to solve the data partitioning problem of variable margins .As a consequence , this paper initially proposes a neighbor-hood homology learning algorithm through using the monomorphic division theory in homology algebra .The neigh-borhood homology learning algorithm reasearchs the margin partitioning problem from the perspective of machine learning.The neighborhood homology learning algorithm includes the method of structuring the neighborhood com -plex, and the criterion for judging the similarity between two given graphs .Finally, this algorithm is justified through the experimental results contrasted with SVM and TVQ on an image dataset named MPEG 7 CE and a data-base of handwritten digits named USPS _ALL.
