Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Stus_Dissertation2021_final.pdf (19.12 MB)

ETD Abstract Container

Abstract Header

Novel Instances and Applications of Shared Knowledge in Computer Vision and Machine Learning Systems

Synakowski, Stuart R
http://orcid.org/0000-0002-5101-1709
http://rave.ohiolink.edu/etdc/view?acc_num=osu1638321951420709

Abstract Details

2021, Doctor of Philosophy, Ohio State University, Electrical and Computer Engineering.
The fields of computer vision and machine learning have made enormous strides in developing models which solve tasks only humans have been capable of solving. However, the models constructed to solve these tasks came at an enormous price in terms of computational resources and data collection. Motivated by the sustainability of continually developing models from scratch to tackle every additional task humans can solve, researchers are interested in efficiently constructing new models for developing solutions to new tasks. The sub-fields of machine learning devoted to this line of research go by many names. Such names include multi-task learning, transfer learning, and few-shot learning. All of these frameworks use the same assumption that knowledge should be shared across models to solve a set of tasks. We define knowledge as the set of conditions used to construct a model that solves a given task. By shared knowledge, we are referring to conditions that are consistently used to construct a set of models which solve a set of tasks. In this work, we address two sets of tasks posed in the fields of computer vision and machine learning. While solving each of these sets of tasks, we show how each of our methods exhibits a novel implementation of shared knowledge leading to many implications for future work in developing systems that further emulate the abilities of human beings. The first set of tasks fall within the sub-field of action analysis, specifically the recognition of intent. Instead of a data-driven approach, we construct a hand-crafted model to infer between intentional/non-intentional movement using common knowledge concepts known by humans. These knowledge concepts are ultimately used to construct an unsupervised method to infer between intentional and non-intentional movement across levels of abstraction. By layers of abstraction we mean that the model needed to solve the most abstract instances of intent recognition, is useful in developing models which solve more tangible instances of intent recognition in the real world. While solving these tasks related to the intention of agents, we show that some instances of shared knowledge do not need to be implemented in a learning framework like in previous methods. We believe this will have many implications in the future development of interpretable systems with common sense reasoning. The second set of tasks fall into the sub-field of image classification using Deep Neural Networks (DNNs). In this work, we address the challenge of training DNNs using fewer training samples. Rather than using previous knowledge-sharing frameworks, we leverage insight regarding the structure of learning in the DNNs themselves, namely, DNNs elicit consistent topological structures when they perform well on image classification tasks. In this work, we provide frameworks to apply this knowledge when solving new image classification tasks using DNNs. Three use cases include performance estimation without a testing set, inferring task-similarity for pre-trained model selection to fine-tune, and training models with fewer training samples via a topological meta-learning strategy. We believe that this work has many implications for future deep learning researchers not only because it applies the insight gained in understanding the structure of DNNs, but it provides a novel and more convenient mechanism to share knowledge. We conclude by discussing additional areas of exploration for the aforementioned methods, along with a brief forecast of the general directions of AI research regarding shared knowledge.
Aleix Martinez (Advisor)
Abhishek Gupta (Committee Member)
Yingbin Liang (Committee Member)
Artificial Intelligence; Computer Engineering; Computer Science
Machine Learning, Computer Vision, Higher-Level Vision, Theory of Mind, Cognitive Science, Deep Learning, Deep Neural Networks, Task-similarity, Meta-learning, Topological Data Analysis, Image Processing, Action Analysis

Recommended Citations

Citations

  • Synakowski, S. R. (2021). Novel Instances and Applications of Shared Knowledge in Computer Vision and Machine Learning Systems [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1638321951420709

    APA Style (7th edition)

  • Synakowski, Stuart. Novel Instances and Applications of Shared Knowledge in Computer Vision and Machine Learning Systems. 2021. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1638321951420709.

    MLA Style (8th edition)

  • Synakowski, Stuart. "Novel Instances and Applications of Shared Knowledge in Computer Vision and Machine Learning Systems." Doctoral dissertation, Ohio State University, 2021. http://rave.ohiolink.edu/etdc/view?acc_num=osu1638321951420709

    Chicago Manual of Style (17th edition)

osu1638321951420709
143
© 2021, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.