서브메뉴
검색
Topics in Artificial Intelligence
Topics in Artificial Intelligence
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152139
- ISBN
- 9798384019435
- DDC
- 658
- 서명/저자
- Topics in Artificial Intelligence
- 발행사항
- [Sl] : Northwestern University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 102 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-02, Section: B.
- 주기사항
- Advisor: Klabjan, Diego.
- 학위논문주기
- Thesis (Ph.D.)--Northwestern University, 2024.
- 초록/해제
- 요약Artificial intelligence (AI) is as a collective term that encompasses various technologies such as machine learning, deep learning, computer vision, and natural language processing, among others. It refers to the capability of computer systems to perform tasks traditionally demanding human intellect and continues to be a vigorously evolving field of research and ongoing advancement. As we navigate through the era of big data, the remarkable advancements in computing power have unlocked new horizons for AI, enabling it to integrate into various aspects of our daily lives, drive innovation and redefine the boundaries of what is possible.AI has become a foundational element in fields ranging from healthcare and finance to manufacturing, education, and transportation, demonstrating its potential impact across diverse sectors and its remarkable adaptability. This versatility has been propelled as AI has advanced its proficiency in processing a wider range of data types, such as images, text, video, and tabular datasets. With this exponential growth and integration into critical decision-making processes, the importance of understanding the rationale behind AI's decisions has also become crucial for building trust and ensuring ethical outcomes. This dissertation consists of three chapters, each of which delves into a distinct topic within the realm of AI, 1) Unsupervised Echocardiogram View Detection via Autoencoder-based Representation Learning, 2) Gradient-boosted Based Structured and Unstructured Learning, and 3) Multi-Layer Attention-Based Explainability via Transformers for Tabular Data.In the first chapter, we focus on a healthcare application and introduce a fully unsupervised echocardiographic view detection framework. It leverages convolutional autoencoders to obtain lower dimensional representations and the K-means algorithm for clustering them into view-related groups. Our approach focuses on discriminative patches from echocardiographic frames. Additionally, we propose a trainable inverse average layer to optimize decoding of average operations. By integrating both public and proprietary datasets, we demonstrate that significant improvements in model performance can be achieved when supplementing proprietary datasets with additional data sources.The second chapter delves into the intricacies of handling multi-modal data. We propose two frameworks to deal with problem settings in which both structured1 and unstructured data are available. Historically, structured data problems are best solved by traditional machine learning models such as boosting and tree-based algorithms, whereas deep learning has been widely applied to problems dealing with images, text, audio, and other unstructured data sources. For the setting in which structured and unstructured data are be available, our proposed frameworks allows joint learning on both data types by integrating the paradigms of boosting models and deep neural networks.The first framework, the boosted-feature-vector deep learning network, learns features from the structured data using gradient boosting and combines them with embeddings from unstructured data via a two-branch deep neural network. The second framework, the two-weak-learner boosting framework, extends the boosting paradigm to the setting with two input data sources. We present and compare first- and second-order methods of this framework.In the third chapter, model explainability is the main emphasis. We propose a graph-oriented attention-based explainability method for tabular data. A transformer architecture for tabular data, which is amenable to explainability, is considered. A novel method to leverage self-attention mechanism to provide explanations by taking into account the attention matrices of all heads and layers as a whole is introduced. The matrices are mapped to a graph structure where groups of features correspond to nodes and attention values to arcs. By finding the maximum probability paths in the graph, we identify groups of features providing larger contributions to explain the model's predictions. To assess the quality of multi-layer attention-based explanations, we compare them with popular attention-, gradient-, and perturbation-based explainability methods.
- 일반주제명
- Industrial engineering
- 일반주제명
- Computer science
- 일반주제명
- Medical imaging
- 일반주제명
- Information technology
- 키워드
- Machine learning
- 키워드
- Deep learning
- 키워드
- Echocardiograms
- 기타저자
- Northwestern University Industrial Engineering and Management Sciences
- 기본자료저록
- Dissertations Abstracts International. 86-02B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017163135
■00520250211152139
■006m o d
■007cr#unu||||||||
■020 ▼a9798384019435
■035 ▼a(MiAaPQ)AAI31484544
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a658
■1001 ▼aTrevino Gavito, Andrea.
■24510▼aTopics in Artificial Intelligence
■260 ▼a[Sl]▼bNorthwestern University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a102 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-02, Section: B.
■500 ▼aAdvisor: Klabjan, Diego.
■5021 ▼aThesis (Ph.D.)--Northwestern University, 2024.
■520 ▼aArtificial intelligence (AI) is as a collective term that encompasses various technologies such as machine learning, deep learning, computer vision, and natural language processing, among others. It refers to the capability of computer systems to perform tasks traditionally demanding human intellect and continues to be a vigorously evolving field of research and ongoing advancement. As we navigate through the era of big data, the remarkable advancements in computing power have unlocked new horizons for AI, enabling it to integrate into various aspects of our daily lives, drive innovation and redefine the boundaries of what is possible.AI has become a foundational element in fields ranging from healthcare and finance to manufacturing, education, and transportation, demonstrating its potential impact across diverse sectors and its remarkable adaptability. This versatility has been propelled as AI has advanced its proficiency in processing a wider range of data types, such as images, text, video, and tabular datasets. With this exponential growth and integration into critical decision-making processes, the importance of understanding the rationale behind AI's decisions has also become crucial for building trust and ensuring ethical outcomes. This dissertation consists of three chapters, each of which delves into a distinct topic within the realm of AI, 1) Unsupervised Echocardiogram View Detection via Autoencoder-based Representation Learning, 2) Gradient-boosted Based Structured and Unstructured Learning, and 3) Multi-Layer Attention-Based Explainability via Transformers for Tabular Data.In the first chapter, we focus on a healthcare application and introduce a fully unsupervised echocardiographic view detection framework. It leverages convolutional autoencoders to obtain lower dimensional representations and the K-means algorithm for clustering them into view-related groups. Our approach focuses on discriminative patches from echocardiographic frames. Additionally, we propose a trainable inverse average layer to optimize decoding of average operations. By integrating both public and proprietary datasets, we demonstrate that significant improvements in model performance can be achieved when supplementing proprietary datasets with additional data sources.The second chapter delves into the intricacies of handling multi-modal data. We propose two frameworks to deal with problem settings in which both structured1 and unstructured data are available. Historically, structured data problems are best solved by traditional machine learning models such as boosting and tree-based algorithms, whereas deep learning has been widely applied to problems dealing with images, text, audio, and other unstructured data sources. For the setting in which structured and unstructured data are be available, our proposed frameworks allows joint learning on both data types by integrating the paradigms of boosting models and deep neural networks.The first framework, the boosted-feature-vector deep learning network, learns features from the structured data using gradient boosting and combines them with embeddings from unstructured data via a two-branch deep neural network. The second framework, the two-weak-learner boosting framework, extends the boosting paradigm to the setting with two input data sources. We present and compare first- and second-order methods of this framework.In the third chapter, model explainability is the main emphasis. We propose a graph-oriented attention-based explainability method for tabular data. A transformer architecture for tabular data, which is amenable to explainability, is considered. A novel method to leverage self-attention mechanism to provide explanations by taking into account the attention matrices of all heads and layers as a whole is introduced. The matrices are mapped to a graph structure where groups of features correspond to nodes and attention values to arcs. By finding the maximum probability paths in the graph, we identify groups of features providing larger contributions to explain the model's predictions. To assess the quality of multi-layer attention-based explanations, we compare them with popular attention-, gradient-, and perturbation-based explainability methods.
■590 ▼aSchool code: 0163.
■650 4▼aIndustrial engineering
■650 4▼aComputer science
■650 4▼aMedical imaging
■650 4▼aInformation technology
■653 ▼aMachine learning
■653 ▼aDeep learning
■653 ▼aNatural language processing
■653 ▼aEchocardiograms
■653 ▼aUnstructured learning
■690 ▼a0546
■690 ▼a0800
■690 ▼a0984
■690 ▼a0489
■690 ▼a0574
■71020▼aNorthwestern University▼bIndustrial Engineering and Management Sciences.
■7730 ▼tDissertations Abstracts International▼g86-02B.
■790 ▼a0163
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17163135▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


