Abstract: This talk discusses the Pareto based approach to solving various machine learning problems. Although machine learning problems inherently have multiple objectives to optimize, these objectives are usually aggregated into a scalar objective optimization function so that traditional mathematical programing techniques such as the gradient based method can be applied. In this talk, we present ideas for solving a range of machine learning problems using evolutionary multi-objective optimization, including model selection and regularization, rule extraction, clustering, feature selection and ensemble generation. We suggest that the multi-objective approach to machine learning may create new perspectives in machine learning and opens up a new avenue for solving machine learning problems.
金耀初 (Yaochu Jin) 分别于1988、1991和1996年在浙江大学电机系获学士、硕士及博士学位，并于2001年在德国波鸿鲁尔大学获工程博士学位。目前为英国萨里大学计算科学系“计算智能”首席教授，“自然计算与应用”研究组主任，萨里大学“数学与计算生物学中心”共同负责人。金耀初是“奖励计划”讲座教授，芬兰国家技术创新局“芬兰讲座教授”，IEEE Fellow。目前担任IEEE Transactions on Cognitive and Developmental Systems主编，Complex & Intelligent Systems主编。曾任IEEE计算智能学会副主席，Distinguished Lecturers Program 杰出讲师。主要研究领域为进化优化，认知与发育系统，生物信息学。