Theses and Dissertations
This collection includes both ASU Theses and Dissertations, submitted by graduate students, and the Barrett, Honors College theses submitted by undergraduate students.
Displaying 1 - 7 of 7
Filtering by
- Creators: Li, Baoxin
Description
With the emergence of edge computing paradigm, many applications such as image recognition and augmented reality require to perform machine learning (ML) and artificial intelligence (AI) tasks on edge devices. Most AI and ML models are large and computational heavy, whereas edge devices are usually equipped with limited computational and storage resources. Such models can be compressed and reduced in order to be placed on edge devices, but they may loose their capability and may not generalize and perform well compared to large models. Recent works used knowledge transfer techniques to transfer information from a large network (termed teacher) to a small one (termed student) in order to improve the performance of the latter. This approach seems to be promising for learning on edge devices, but a thorough investigation on its effectiveness is lacking.
The purpose of this work is to provide an extensive study on the performance (both in terms of accuracy and convergence speed) of knowledge transfer, considering different student-teacher architectures, datasets and different techniques for transferring knowledge from teacher to student.
A good performance improvement is obtained by transferring knowledge from both the intermediate layers and last layer of the teacher to a shallower student. But other architectures and transfer techniques do not fare so well and some of them even lead to negative performance impact. For example, a smaller and shorter network, trained with knowledge transfer on Caltech 101 achieved a significant improvement of 7.36\% in the accuracy and converges 16 times faster compared to the same network trained without knowledge transfer. On the other hand, smaller network which is thinner than the teacher network performed worse with an accuracy drop of 9.48\% on Caltech 101, even with utilization of knowledge transfer.
The purpose of this work is to provide an extensive study on the performance (both in terms of accuracy and convergence speed) of knowledge transfer, considering different student-teacher architectures, datasets and different techniques for transferring knowledge from teacher to student.
A good performance improvement is obtained by transferring knowledge from both the intermediate layers and last layer of the teacher to a shallower student. But other architectures and transfer techniques do not fare so well and some of them even lead to negative performance impact. For example, a smaller and shorter network, trained with knowledge transfer on Caltech 101 achieved a significant improvement of 7.36\% in the accuracy and converges 16 times faster compared to the same network trained without knowledge transfer. On the other hand, smaller network which is thinner than the teacher network performed worse with an accuracy drop of 9.48\% on Caltech 101, even with utilization of knowledge transfer.
ContributorsSistla, Ragini (Author) / Zhao, Ming (Thesis advisor, Committee member) / Li, Baoxin (Committee member) / Tong, Hanghang (Committee member) / Arizona State University (Publisher)
Created2018
Description
Social media refers computer-based technology that allows the sharing of information and building the virtual networks and communities. With the development of internet based services and applications, user can engage with social media via computer and smart mobile devices. In recent years, social media has taken the form of different activities such as social network, business network, text sharing, photo sharing, blogging, etc. With the increasing popularity of social media, it has accumulated a large amount of data which enables understanding the human behavior possible. Compared with traditional survey based methods, the analysis of social media provides us a golden opportunity to understand individuals at scale and in turn allows us to design better services that can tailor to individuals’ needs. From this perspective, we can view social media as sensors, which provides online signals from a virtual world that has no geographical boundaries for the real world individual's activity.
One of the key features for social media is social, where social media users actively interact to each via generating content and expressing the opinions, such as post and comment in Facebook. As a result, sentiment analysis, which refers a computational model to identify, extract or characterize subjective information expressed in a given piece of text, has successfully employs user signals and brings many real world applications in different domains such as e-commerce, politics, marketing, etc. The goal of sentiment analysis is to classify a user’s attitude towards various topics into positive, negative or neutral categories based on textual data in social media. However, recently, there is an increasing number of people start to use photos to express their daily life on social media platforms like Flickr and Instagram. Therefore, analyzing the sentiment from visual data is poise to have great improvement for user understanding.
In this dissertation, I study the problem of understanding human sentiments from large scale collection of social images based on both image features and contextual social network features. We show that neither
visual features nor the textual features are by themselves sufficient for accurate sentiment prediction. Therefore, we provide a way of using both of them, and formulate sentiment prediction problem in two scenarios: supervised and unsupervised. We first show that the proposed framework has flexibility to incorporate multiple modalities of information and has the capability to learn from heterogeneous features jointly with sufficient training data. Secondly, we observe that negative sentiment may related to human mental health issues. Based on this observation, we aim to understand the negative social media posts, especially the post related to depression e.g., self-harm content. Our analysis, the first of its kind, reveals a number of important findings. Thirdly, we extend the proposed sentiment prediction task to a general multi-label visual recognition task to demonstrate the methodology flexibility behind our sentiment analysis model.
One of the key features for social media is social, where social media users actively interact to each via generating content and expressing the opinions, such as post and comment in Facebook. As a result, sentiment analysis, which refers a computational model to identify, extract or characterize subjective information expressed in a given piece of text, has successfully employs user signals and brings many real world applications in different domains such as e-commerce, politics, marketing, etc. The goal of sentiment analysis is to classify a user’s attitude towards various topics into positive, negative or neutral categories based on textual data in social media. However, recently, there is an increasing number of people start to use photos to express their daily life on social media platforms like Flickr and Instagram. Therefore, analyzing the sentiment from visual data is poise to have great improvement for user understanding.
In this dissertation, I study the problem of understanding human sentiments from large scale collection of social images based on both image features and contextual social network features. We show that neither
visual features nor the textual features are by themselves sufficient for accurate sentiment prediction. Therefore, we provide a way of using both of them, and formulate sentiment prediction problem in two scenarios: supervised and unsupervised. We first show that the proposed framework has flexibility to incorporate multiple modalities of information and has the capability to learn from heterogeneous features jointly with sufficient training data. Secondly, we observe that negative sentiment may related to human mental health issues. Based on this observation, we aim to understand the negative social media posts, especially the post related to depression e.g., self-harm content. Our analysis, the first of its kind, reveals a number of important findings. Thirdly, we extend the proposed sentiment prediction task to a general multi-label visual recognition task to demonstrate the methodology flexibility behind our sentiment analysis model.
ContributorsWang, Yilin (Author) / Li, Baoxin (Thesis advisor) / Liu, Huan (Committee member) / Tong, Hanghang (Committee member) / Chang, Yi (Committee member) / Arizona State University (Publisher)
Created2018
Description
High-level inference tasks in video applications such as recognition, video retrieval, and zero-shot classification have become an active research area in recent years. One fundamental requirement for such applications is to extract high-quality features that maintain high-level information in the videos.
Many video feature extraction algorithms have been purposed, such as STIP, HOG3D, and Dense Trajectories. These algorithms are often referred to as “handcrafted” features as they were deliberately designed based on some reasonable considerations. However, these algorithms may fail when dealing with high-level tasks or complex scene videos. Due to the success of using deep convolution neural networks (CNNs) to extract global representations for static images, researchers have been using similar techniques to tackle video contents. Typical techniques first extract spatial features by processing raw images using deep convolution architectures designed for static image classifications. Then simple average, concatenation or classifier-based fusion/pooling methods are applied to the extracted features. I argue that features extracted in such ways do not acquire enough representative information since videos, unlike images, should be characterized as a temporal sequence of semantically coherent visual contents and thus need to be represented in a manner considering both semantic and spatio-temporal information.
In this thesis, I propose a novel architecture to learn semantic spatio-temporal embedding for videos to support high-level video analysis. The proposed method encodes video spatial and temporal information separately by employing a deep architecture consisting of two channels of convolutional neural networks (capturing appearance and local motion) followed by their corresponding Fully Connected Gated Recurrent Unit (FC-GRU) encoders for capturing longer-term temporal structure of the CNN features. The resultant spatio-temporal representation (a vector) is used to learn a mapping via a Fully Connected Multilayer Perceptron (FC-MLP) to the word2vec semantic embedding space, leading to a semantic interpretation of the video vector that supports high-level analysis. I evaluate the usefulness and effectiveness of this new video representation by conducting experiments on action recognition, zero-shot video classification, and semantic video retrieval (word-to-video) retrieval, using the UCF101 action recognition dataset.
Many video feature extraction algorithms have been purposed, such as STIP, HOG3D, and Dense Trajectories. These algorithms are often referred to as “handcrafted” features as they were deliberately designed based on some reasonable considerations. However, these algorithms may fail when dealing with high-level tasks or complex scene videos. Due to the success of using deep convolution neural networks (CNNs) to extract global representations for static images, researchers have been using similar techniques to tackle video contents. Typical techniques first extract spatial features by processing raw images using deep convolution architectures designed for static image classifications. Then simple average, concatenation or classifier-based fusion/pooling methods are applied to the extracted features. I argue that features extracted in such ways do not acquire enough representative information since videos, unlike images, should be characterized as a temporal sequence of semantically coherent visual contents and thus need to be represented in a manner considering both semantic and spatio-temporal information.
In this thesis, I propose a novel architecture to learn semantic spatio-temporal embedding for videos to support high-level video analysis. The proposed method encodes video spatial and temporal information separately by employing a deep architecture consisting of two channels of convolutional neural networks (capturing appearance and local motion) followed by their corresponding Fully Connected Gated Recurrent Unit (FC-GRU) encoders for capturing longer-term temporal structure of the CNN features. The resultant spatio-temporal representation (a vector) is used to learn a mapping via a Fully Connected Multilayer Perceptron (FC-MLP) to the word2vec semantic embedding space, leading to a semantic interpretation of the video vector that supports high-level analysis. I evaluate the usefulness and effectiveness of this new video representation by conducting experiments on action recognition, zero-shot video classification, and semantic video retrieval (word-to-video) retrieval, using the UCF101 action recognition dataset.
ContributorsHu, Sheng-Hung (Author) / Li, Baoxin (Thesis advisor) / Turaga, Pavan (Committee member) / Liang, Jianming (Committee member) / Tong, Hanghang (Committee member) / Arizona State University (Publisher)
Created2016
Description
Visual Question Answering (VQA) is a new research area involving technologies ranging from computer vision, natural language processing, to other sub-fields of artificial intelligence such as knowledge representation. The fundamental task is to take as input one image and one question (in text) related to the given image, and to generate a textual answer to the input question. There are two key research problems in VQA: image understanding and the question answering. My research mainly focuses on developing solutions to support solving these two problems.
In image understanding, one important research area is semantic segmentation, which takes images as input and output the label of each pixel. As much manual work is needed to label a useful training set, typical training sets for such supervised approaches are always small. There are also approaches with relaxed labeling requirement, called weakly supervised semantic segmentation, where only image-level labels are needed. With the development of social media, there are more and more user-uploaded images available
on-line. Such user-generated content often comes with labels like tags and may be coarsely labelled by various tools. To use these information for computer vision tasks, I propose a new graphic model by considering the neighborhood information and their interactions to obtain the pixel-level labels of the images with only incomplete image-level labels. The method was evaluated on both synthetic and real images.
In question answering, my research centers on best answer prediction, which addressed two main research topics: feature design and model construction. In the feature design part, most existing work discussed how to design effective features for answer quality / best answer prediction. However, little work mentioned how to design features by considering the relationship between answers of one given question. To fill this research gap, I designed new features to help improve the prediction performance. In the modeling part, to employ the structure of the feature space, I proposed an innovative learning-to-rank model by considering the hierarchical lasso. Experiments with comparison with the state-of-the-art in the best answer prediction literature have confirmed
that the proposed methods are effective and suitable for solving the research task.
In image understanding, one important research area is semantic segmentation, which takes images as input and output the label of each pixel. As much manual work is needed to label a useful training set, typical training sets for such supervised approaches are always small. There are also approaches with relaxed labeling requirement, called weakly supervised semantic segmentation, where only image-level labels are needed. With the development of social media, there are more and more user-uploaded images available
on-line. Such user-generated content often comes with labels like tags and may be coarsely labelled by various tools. To use these information for computer vision tasks, I propose a new graphic model by considering the neighborhood information and their interactions to obtain the pixel-level labels of the images with only incomplete image-level labels. The method was evaluated on both synthetic and real images.
In question answering, my research centers on best answer prediction, which addressed two main research topics: feature design and model construction. In the feature design part, most existing work discussed how to design effective features for answer quality / best answer prediction. However, little work mentioned how to design features by considering the relationship between answers of one given question. To fill this research gap, I designed new features to help improve the prediction performance. In the modeling part, to employ the structure of the feature space, I proposed an innovative learning-to-rank model by considering the hierarchical lasso. Experiments with comparison with the state-of-the-art in the best answer prediction literature have confirmed
that the proposed methods are effective and suitable for solving the research task.
ContributorsTian, Qiongjie (Author) / Li, Baoxin (Thesis advisor) / Tong, Hanghang (Committee member) / Davulcu, Hasan (Committee member) / Yang, Yezhou (Committee member) / Arizona State University (Publisher)
Created2017
Description
Graph is a ubiquitous data structure, which appears in a broad range of real-world scenarios. Accordingly, there has been a surge of research to represent and learn from graphs in order to accomplish various machine learning and graph analysis tasks. However, most of these efforts only utilize the graph structure while nodes in real-world graphs usually come with a rich set of attributes. Typical examples of such nodes and their attributes are users and their profiles in social networks, scientific articles and their content in citation networks, protein molecules and their gene sets in biological networks as well as web pages and their content on the Web. Utilizing node features in such graphs---attributed graphs---can alleviate the graph sparsity problem and help explain various phenomena (e.g., the motives behind the formation of communities in social networks). Therefore, further study of attributed graphs is required to take full advantage of node attributes.
In the wild, attributed graphs are usually unlabeled. Moreover, annotating data is an expensive and time-consuming process, which suffers from many limitations such as annotators’ subjectivity, reproducibility, and consistency. The challenges of data annotation and the growing increase of unlabeled attributed graphs in various real-world applications significantly demand unsupervised learning for attributed graphs.
In this dissertation, I propose a set of novel models to learn from attributed graphs in an unsupervised manner. To better understand and represent nodes and communities in attributed graphs, I present different models in node and community levels. In node level, I utilize node features as well as the graph structure in attributed graphs to learn distributed representations of nodes, which can be useful in a variety of downstream machine learning applications. In community level, with a focus on social media, I take advantage of both node attributes and the graph structure to discover not only communities but also their sentiment-driven profiles and inter-community relations (i.e., alliance, antagonism, or no relation). The discovered community profiles and relations help to better understand the structure and dynamics of social media.
In the wild, attributed graphs are usually unlabeled. Moreover, annotating data is an expensive and time-consuming process, which suffers from many limitations such as annotators’ subjectivity, reproducibility, and consistency. The challenges of data annotation and the growing increase of unlabeled attributed graphs in various real-world applications significantly demand unsupervised learning for attributed graphs.
In this dissertation, I propose a set of novel models to learn from attributed graphs in an unsupervised manner. To better understand and represent nodes and communities in attributed graphs, I present different models in node and community levels. In node level, I utilize node features as well as the graph structure in attributed graphs to learn distributed representations of nodes, which can be useful in a variety of downstream machine learning applications. In community level, with a focus on social media, I take advantage of both node attributes and the graph structure to discover not only communities but also their sentiment-driven profiles and inter-community relations (i.e., alliance, antagonism, or no relation). The discovered community profiles and relations help to better understand the structure and dynamics of social media.
ContributorsSalehi, Amin (Author) / Davulcu, Hasan (Thesis advisor) / Liu, Huan (Committee member) / Li, Baoxin (Committee member) / Tong, Hanghang (Committee member) / Arizona State University (Publisher)
Created2019
Description
Machine learning (ML) and deep neural networks (DNNs) have achieved great success in a variety of application domains, however, despite significant effort to make these networks robust, they remain vulnerable to adversarial attacks in which input that is perceptually indistinguishable from natural data can be erroneously classified with high prediction confidence. Works on defending against adversarial examples can be broadly classified as correcting or detecting, which aim, respectively at negating the effects of the attack and correctly classifying the input, or detecting and rejecting the input as adversarial. In this work, a new approach for detecting adversarial examples is proposed. The approach takes advantage of the robustness of natural images to noise. As noise is added to a natural image, the prediction probability of its true class drops, but the drop is not sudden or precipitous. The same seems to not hold for adversarial examples. In other word, the stress response profile for natural images seems different from that of adversarial examples, which could be detected by their stress response profile. An evaluation of this approach for detecting adversarial examples is performed on the MNIST, CIFAR-10 and ImageNet datasets. Experimental data shows that this approach is effective at detecting some adversarial examples on small scaled simple content images and with little sacrifice on benign accuracy.
ContributorsSun, Lin (Author) / Bazzi, Rida (Thesis advisor) / Li, Baoxin (Committee member) / Tong, Hanghang (Committee member) / Arizona State University (Publisher)
Created2019
Description
Data mining, also known as big data analysis, has been identified as a critical and challenging process for a variety of applications in real-world problems. Numerous datasets are collected and generated every day to store the information. The rise in the number of data volumes and data modality has resulted in the increased demand for data mining methods and strategies of finding anomalies, patterns, and correlations within large data sets to predict outcomes. Effective machine learning methods are widely adapted to build the data mining pipeline for various purposes like business understanding, data understanding, data preparation, modeling, evaluation, and deployment. The major challenges for effectively and efficiently mining big data include (1) data heterogeneity and (2) missing data. Heterogeneity is the natural characteristic of big data, as the data is typically collected from different sources with diverse formats. The missing value is the most common issue faced by the heterogeneous data analysis, which resulted from variety of factors including the data collecting processing, user initiatives, erroneous data entries, and so on. In response to these challenges, in this thesis, three main research directions with application scenarios have been investigated: (1) Mining and Formulating Heterogeneous Data, (2) missing value imputation strategy in various application scenarios in both offline and online manner, and (3) missing value imputation for multi-modality data. Multiple strategies with theoretical analysis are presented, and the evaluation of the effectiveness of the proposed algorithms compared with state-of-the-art methods is discussed.
Contributorsliu, Xu (Author) / He, Jingrui (Thesis advisor) / Xue, Guoliang (Thesis advisor) / Li, Baoxin (Committee member) / Tong, Hanghang (Committee member) / Arizona State University (Publisher)
Created2021