Consistent and coherent learning with δ-delay
Consistent and coherent learning with δ-delay
复制标题
DOI：
10.1016/j.ic.2008.06.005
复制
发表时间：
2008-11-01
影响因子：
1
通讯作者：
Zeugmann, Thomas
中科院分区：
文献类型：
--
作者：
Akama, Yohji;Zeugmann, Thomas
一个一致的学习器需要正确和完整地反映到目前为止收到的所有数据在其实际假设。虽然这个要求听起来很合理，但它可能导致学习问题的不可解决性。因此，在本文中，一致性学习的几个变化进行了介绍和研究。这些变化允许所谓的增量延迟，将一致性要求放宽到除了最后的增量数据之外的所有数据。此外，我们引入了相干学习的概念（再次具有增量延迟），要求学习者仅正确反映所看到的最后一个数据（仅第n个增量数据）。我们的结果是多方面的。首先，我们提供了一致的学习与delta-delay的复杂性和可计算的编号方面的特征。其次，我们建立严格的层次结构，所有一致的学习模型与delta-delay依赖于delta。最后，它表明，所有的一致性学习模型与delta-delay是完全一样强大的，其相应的一致性学习模型与delta-delay。(c)2008年爱思唯尔公司版权所有
A consistent learner is required to correctly and completely reflect in its actual hypothesis all data received so far. Though this demand sounds quite plausible, it may lead to the unsolvability of the learning problem. Therefore, in the present paper several variations of consistent learning are introduced and studied. These variations allow a so-called delta-delay relaxing the consistency demand to all but the last delta data. Additionally, we introduce the notion of coherent learning (again with delta-delay) requiring the learner to correctly reflect only the last datum (only the n - delta th datum) seen. Our results are manyfold. First, we provide characterizations for consistent learning with delta-delay in terms of complexity and computable numberings. Second, we establish strict hierarchies for all consistent learning models with delta-delay in dependence on delta. Finally, it is shown that all models of coherent learning with delta-delay are exactly as powerful as their corresponding consistent learning models with delta-delay. (c) 2008 Elsevier Inc. All rights reserved