Research of Sequential Multitask Learning and ItsApplication to Patlern Recognition

本课题针对多任务模式识别问题开发了一种新的学习算法。该项目考虑在线学习多个分类任务，其中没有提供关于训练样例的任务类别的信息。因此，该算法需要具有自动任务识别能力，以正确学习不同的分类任务。学习模式是“在线”的，不同任务的训练样例以随机方式混合，依次给出。在“在线训练”过程中，假设分类任务是相互关联的，并且它们的训练样例以随机序列出现。因此，当训练样例转换为不同的任务时，学习算法必须不断地从学习一个任务切换到另一个任务。这也意味着学习算法必须自动、快速地检测任务的变化，并利用以前任务的知识来学习新的任务。总的来说，自动任务识别属于无监督学习的范畴，因为没有提供训练样例的任务类别信息给算法。使用几个人工生成的和三个UCI数据集对算法的性能进行了评估。本课题的实验验证了所提出的算法确实能够获取和积累任务知识，并且已经学习到的任务知识的迁移提高了新任务的知识获取速度。

This research project developed a new learning algorithm for the multi-task pattern recognition problem. This project considers learning multiple classification tasks online where no information is ever provided about the task category of a training example. The algorithm thus needs an automated task recognition capability to properly learn the different classification tasks. The learning mode is “online" where training examples for different tasks are mixed in a random fashion and given sequentially one after another. It is assumed that the classification tasks are related to each other and that their training examples appear in random sequences during “online training." Thus, the learning algorithm has to continually switch from learning one task to another whenever the training examples change to a different task. This also implies that the learning algorithm has to detect task changes automatically and fast and utilize knowledge of previous tasks to learn new tasks. Overall, automated task recognition falls in the category of unsupervised learning since no information about task categories of training examples is provided to the algorithm. The performance of the algorithm is evaluated using several artificially generated and three UCI datasets. The experiments in this project verify that the proposed algorithm can indeed acquire and accumulate task knowledge and that the transfer of knowledge from tasks already learned enhances the speed of knowledge acquisition on new tasks.