Multitask Learning is a learning framework which explores the concept of sharing training information among multiple related tasks to improve the generalization error of each task. The benefits of multitask learning have been shown both empirically and theoretically. There are a number of fields that benefit from multitask learning such as toxicology, image annotation, compressive sensing etc. However, majority of multitask learning algorithms make a very important key assumption that all the tasks are related to each other in a similar fashion in multitask learning. The users often do not have the knowledge of which tasks are related and train all tasks together. This results in sharing of training information even among the unrelated tasks. Training unrelated tasks together can cause a negative transfer and deteriorate the performance of multitask learning. For example, consider the case of predicting in vivo toxicity of chemicals at various endpoints from the chemical structure. Toxicity at all the endpoints are not related. Since, biological networks are highly complex, it is also not possible to predetermine which endpoints are related. Training all the endpoints together may cause a negative effect on the overall performance. Therefore, it is important to establish the task relationship models in multitask learning.
Multitask learning with task relationship modeling may be explored in three different settings, namely, static learning, online fixed task learning and most recent lifelong learning. The multitask learning algorithms in static setting have been present for more than a decade and there is a lot of literature in this field. However, utilization of task relationships in multitask learning framework has been studied in detail for past several years only. The literature which uses feature selection with task relationship modeling is even further limited.
For the cases of online and lifelong learning, task relationship modeling becomes a challenge. In online learning, the knowledge of all the tasks is present before starting the training of the algorithms, and the samples arrive in online fashion. However, in case of lifelong multitask learning, the tasks also arrive in an online fashion. Therefore, modeling the task relationship is even a further challenge in lifelong multitask learning framework as compared to online multitask learning.
The main contribution of this thesis is to propose a framework for modeling task relationships in lifelong multitask learning. The initial algorithms are preliminary studies which focus on static setting and learn the clusters of related tasks with feature selection. These algorithms enforce that all the tasks which are related select a common set of features. The later part of the thesis shifts gear to lifelong multitask learning setting. Here, we propose learning functions to represent the relationship between tasks. Learning functions is faster and computationally less expensive as opposed to the traditional manner of learning fixed sized matrices for depicting the task relationship models.
|Commitee:||Agah, Arvin, Chakrabarti, Swapan, Hui, Rongqing, Wang, Zhuo|
|School:||University of Kansas|
|Department:||Electrical Engineering and Computer Science|
|School Location:||United States -- Kansas|
|Source:||DAI-B 77/01(E), Dissertation Abstracts International|
|Subjects:||Computer Engineering, Artificial intelligence, Computer science|
|Keywords:||Feature selection, Lifelong learning, Multitask learning, Online multitask learning, Task space partition|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
The supplemental file or files you are about to download were provided to ProQuest by the author as part of a
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
Copyright of the original materials contained in the supplemental file is retained by the author and your access to the
supplemental files is subject to the ProQuest Terms and Conditions of use.
Depending on the size of the file(s) you are downloading, the system may take some time to download them. Please be