摘要
为研究机器学习的推广误差 ,提出了变一误差估计条件下一种新的学习算法稳定 逐点假设稳定 ,并讨论了逐点假设稳定、CV稳定、重叠稳定以及弱假设稳定四种学习算法稳定之间的关系 ,得出了逐点假设稳定是这四种学习算法稳定中最弱的学习算法稳定的结论。
A new notion of algorithmic stability under the change one error estimates,Pointwise Hypothesis Stability was introduced to study the Generalization error of machine Learning,and the relationships of four kinds of learning algorithmic stabilities was also discussed in the paper.The four kinds of learning algorithmic stabilities are Pointwise Hypothesis Stability,Cross Validation Stability,Overlap Stability,Weakly Hypothesis Stability.From these relationships,we can find that Pointwise Hypothesis Stability is the most weakly learning algorithmic stability of the four learning algorithmic stabilities.
出处
《计算机应用与软件》
CSCD
北大核心
2005年第1期30-31,111,共3页
Computer Applications and Software
基金
湖北省自然科学基金资助项目 (99J1 69)。
