论文部分内容阅读
用户适应问题是智能化人机接口设计中的一个重点和难点 ,它为人机接口的人性化、智能化和个性化提供支持 .由于用户自身的特殊性 ,例如笔迹、口音、绘画习惯等 ,系统很难同时适应多个用户的使用 .支持向量机 (SVM)的学习方法是基于小样本的学习方法 ,它实现了结构风险最小化 ,避免了在学习过程中存在的过学习现象 .增量学习能有效地利用历史训练结果 ,从而能在很小的时间空间代价下实现新样本的学习 .基于SVM增量学习的方法 ,能从用户的历史数据中找到根本特性 ,而不会将可能造成用户冲突的、特定用户习惯的特征也记录下来 ,因此也就不会产生用户冲突 .在在线图形识别系统中 ,与基于规则的用户适应方法相比 ,基于SVM增量学习的方法可以适应多个用户 .对基于SVM的学习方法进行了分析 ,将其与用户适应紧密结合起来 .对涉及到的 3个方面进行了理论和实验上分析和对比 :重复学习和增量学习 ;Syed等提出的和Xiao等提出的两种不同的基于SVM增量学习方法 ;“一对一”和“一对多”两种不同的多值分类构造方法 .从而得出结论 :两种增量学习方法都要明显优于重复学习 ;Syed等提出的增量学习方法在精度和效率都好于Xiao等提出的方法 ;一对一的多值分类构造方法要优于一对多的多值分类构造方法 .
User adaptation is a key and difficult point in the design of intelligent human-machine interface, which provides support for the humanization, intelligence and personalization of the human-machine interface.Because of the user’s own particularity, such as handwriting, accent, drawing habit, etc., the system It is difficult to adapt to the use of multiple users at the same time.Support vector machine (SVM) learning method is based on the small sample learning method, which realizes the structural risk minimization, avoids over learning phenomenon existing in the learning process. Can effectively use the results of historical training, so as to realize the learning of new samples with little time and space cost.The method based on incremental learning of SVM can find the fundamental characteristics from the user’s historical data without the possibility of causing the user Conflicting and user-specific features are also recorded and therefore user conflicts are not generated.In the online graphical recognition system, the method based on SVM incremental learning can accommodate multiple users compared to rule-based user-adaptive methods The SVM-based learning method was analyzed and closely integrated with user adaptation. The three aspects involved Theoretical and experimental analysis and comparison: repetitive learning and incremental learning; two different SVM incremental learning methods proposed by Syed et al. And proposed by Xiao et al; two different types of “one to one” and “one to many” And concludes that both incremental learning methods are obviously better than repetitive learning. The incremental learning method proposed by Syed et al. Is better than the one proposed by Xiao et al. In one-on-one Multi-value classification construction method is better than one to many multi-value classification construction method.