CAJ | 학술논문

推广性能是机器学习理论研究的主要目的之一.为了研究相依序列下采用ERM算法的学习机器的推广性能,本文基于β-混合序列建立了采用ERM算法的学习机器的经验风险到它的期望风险相对一致收敛速率的界.这个界不仅把基于独立序列下已有的结果推广到β-混合相依序列的情况,而且对β-混合相依序列现有的一些结论进行了改进.得到了β-混合相依序列下,采用ERM算法的学习机器的推广性能的界.
추엄성능시궤기학습이론연구적주요목적지일.위료연구상의서렬하채용ERM산법적학습궤기적추엄성능,본문기우β-혼합서렬건립료채용ERM산법적학습궤기적경험풍험도타적기망풍험상대일치수렴속솔적계.저개계불부파기우독립서렬하이유적결과추엄도β-혼합상의서렬적정황,이차대β-혼합상의서렬현유적일사결론진행료개진.득도료β-혼합상의서렬하,채용ERM산법적학습궤기적추엄성능적계.
The generalization performance is the main purpose of machine learning theoretical re-search. To study the generalization ability of ERM algorithms with dependent observa-tions in this paper, we derive the bounds on the rate of relative uniform convergence of theempirical risks to their expected risks with beta-mixing dependent sequences, which extendthe previous results with the independent and identically distributed (i.i.d.) sequence, andimprove current results with beta-mixing dependent sequence. We also establish the boundthat describes the generalization ability of ERM algorithm with beta-mixing sequences.