2026-05-24T19:37:00Zhttps://keep.lib.asu.edu/oai/request

oai:keep.lib.asu.edu:node-1559632024-12-20T18:25:12Zoai_pmh:alloai_pmh:repo_items

155963 https://hdl.handle.net/2286/R.I.46233 http://rightsstatements.org/vocab/InC/1.0/ All Rights Reserved 2017 138 pages Doctoral Dissertation Academic theses Text eng Venkatesan, Ragav Li, Baoxin Turaga, Pavan Yang, Yezhou Davulcu, Hasan Arizona State University Doctoral Dissertation Computer Science 2017 Computer Vision as a eld has gone through signicant changes in the last decade.<br/><br/>The eld has seen tremendous success in designing learning systems with hand-crafted<br/><br/>features and in using representation learning to extract better features. In this dissertation<br/><br/>some novel approaches to representation learning and task learning are studied.<br/><br/>Multiple-instance learning which is generalization of supervised learning, is one<br/><br/>example of task learning that is discussed. In particular, a novel non-parametric k-<br/><br/>NN-based multiple-instance learning is proposed, which is shown to outperform other<br/><br/>existing approaches. This solution is applied to a diabetic retinopathy pathology<br/><br/>detection problem eectively.<br/><br/>In cases of representation learning, generality of neural features are investigated<br/><br/>rst. This investigation leads to some critical understanding and results in feature<br/><br/>generality among datasets. The possibility of learning from a mentor network instead<br/><br/>of from labels is then investigated. Distillation of dark knowledge is used to eciently<br/><br/>mentor a small network from a pre-trained large mentor network. These studies help<br/><br/>in understanding representation learning with smaller and compressed networks. Computer Science Dataset Generality Deep learning Image Representations Mentee Networks Multiple Instance Learning Novel Image Representations and Learning Tasks