Asset Details
Do you wish to reserve the book?
Learning Deep Models Under Constraints of Annotated Data Insufficiency
Hey, we have placed the reservation for you!
By the way, why not check out events that you can attend while you pick your title.
Oops! Something went wrong.
Looks like we were not able to place the reservation. Kindly try again later.
Are you sure you want to remove the book from the shelf?
Learning Deep Models Under Constraints of Annotated Data Insufficiency
Do you wish to request the book?
Learning Deep Models Under Constraints of Annotated Data Insufficiency
Please be aware that the book you have requested cannot be checked out. If you would like to checkout this book, you can reserve another copy
We have requested the book for you!
Your request is successful and it will be processed during the Library working hours. Please check the status of your request in My Requests.
Oops! Something went wrong.
Looks like we were not able to place your request. Kindly try again later.
Dissertation
Learning Deep Models Under Constraints of Annotated Data Insufficiency
2020
Overview
Deep learning and machine learning have been contributing significantly to many fields in the scientific community, from computer vision to natural language processing and robotics. These data driven approaches provide us with powerful tools that accommodate modeling sensory data, such as images, videos and audio. One major drawback that is associated with these models is the large amount of unknown parameters to learn. These many unknown variables are usually learned from expert annotations that accompany data. However, access to annotations is not always straight-forward and it can be costly. In this dissertation we are going to focus on problems where sufficient annotated data are not provided for learning. We will investigate different applications in computer vision such as image segmentation, image classification and regression and propose corresponding constraints that mitigate the consequences associated with annotated data insufficiency. We propose the use of structured output losses as unsupervised loss functions to be jointly learned with supervised loss functions for the task of semantic segmentation and scene labeling. The choice of these unsupervised loss functions is motivated by the application as we observe that the probability maps generated by a segmentation algorithm should be smooth. We further propose to use adversarial training to match the output distributions of a segmentation network trained only by using source annotated data. By matching the output distributions of source and target data we are able to perform domain adaptation for the target data. We also use shape models in conjunction with deep networks to perform segmentation and reconstruction on noisy and corrupted image data. Finally we look at a chronic disease prevalence regression problem where we modify the architecture of traditional approaches to be able to learn regression models with less annotated data. This modification to the architecture of the model will require the network to take sets of inputs rather than single inputs to the network.
Publisher
ProQuest Dissertations & Theses
Subject
ISBN
9798837550676
