자료유형  

 학위논문

Control Number  

0017164424

International Standard Book Number  

9798346387824

Dewey Decimal Classification Number  

620

Main Entry-Personal Name  

Chiu, Cho-Chun.

Publication, Distribution, etc. (Imprint  

[S.l.] : The Pennsylvania State University., 2024

Publication, Distribution, etc. (Imprint  

Ann Arbor : ProQuest Dissertations & Theses, 2024

Physical Description  

173 p.

General Note  

Source: Dissertations Abstracts International, Volume: 86-05, Section: B.

General Note  

Advisor: He, Ting.

Dissertation Note  

Thesis (Ph.D.)--The Pennsylvania State University, 2024.

Summary, Etc.  

요약In the context of this dissertation, learning in network includes to two parts: learning about network and learning over network.The former refers to applying network tomography to learn the link performance metrics (e.q., link delays) from end-to-end measurement. The later refers to training a machine learning model over the network. Our four pieces of work focus on optimizing the designs of network tomography, data curation, and model training.In our first piece of work, we aim to understand the fundamental limit of network tomography in an adversarial environment. We formulate and analyze a novel type of attack that aims at maximally degrading the performance of targeted paths without being localized by network tomography, called stealthy DeGrading of Service (DGoS) attack. By analyzing properties of the optimal attack strategy, we formulate novel combinatorial optimizations to design the optimal attack strategy, which are then linked to well-known NP-hard problems and approximation algorithms. As a byproduct, our algorithms also identify approximations of the most vulnerable set of links that once manipulated, can inflict the maximum performance degradation. Our evaluations on real topologies demonstrate the large potential damage of such attacks, signaling the need of new defenses.In our second piece of work, we aim to design the defense strategy of DGoS attack for network tomography. In order to mitigate the performance degradation, our model allows network tomography to measure a larger set of paths, e.g., by sending probes on some paths not carrying data flows. By developing and analyzing the optimal attack strategy, we quantify the maximum damage of such an attack. We further develop a defense strategy by formulating and solving a Stackelberg game to select the best set of measurement paths under a budget constraint. Our evaluations on real topologies validate the efficacy of the proposed defense strategy while identifying areas for further improvement.In our third piece of work, we aim to reduce the communication cost for decentralized learning. We consider the problem of training a given machine learning model by decentralized parallel stochastic gradient descent over training data distributed across multiple nodes. To optimize communication cost, we propose a framework and efficient algorithms to design the communication patterns through the mixing matrix sampling, which governs not only which nodes should communicate with each other but also what weights the communicated parameters should carry during parameter aggregation. Our framework is designed to minimize the total cost incurred until convergence based on any given cost model that is additive over iterations. Our solution achieves superior performance under a variety of network settings and cost models in experiments based on real datasets and topologies, saving 24-50% of the cost compared to the state-of-the-art design without compromising the quality of the trained model.In our fourth piece of work, we aim to reduce the communication cost for machine learning in wireless body area network (WBAN). We consider the problem of training a target model over the data collected through body sensors in WBAN, where resources at body sensors are limited. Since samples are unlabeled in the data collection phase, classical active learning can not be applied to reduce the communication cost. To handle these challenges, we propose a two-phased active learning method, consisting of an online phase where a coreset construction algorithm is proposed to select a subset of unlabeled samples based on their noisy predictions, and an offline phase where the selected samples are labeled to train the target model. Our evaluation based on real health monitoring data and our own experimentation demonstrates that our solution can drastically save the data curation cost without sacrificing the quality of the target model.

Subject Added Entry-Topical Term  

Wireless networks.

Subject Added Entry-Topical Term  

Tomography.

Subject Added Entry-Topical Term  

Communication.

Subject Added Entry-Topical Term  

Bandwidths.

Subject Added Entry-Topical Term  

Linear programming.

Subject Added Entry-Topical Term  

Defense.

Subject Added Entry-Topical Term  

Energy consumption.

Subject Added Entry-Topical Term  

Electrical engineering.

Subject Added Entry-Topical Term  

Energy.

Subject Added Entry-Topical Term  

Medical imaging.

Added Entry-Corporate Name  

The Pennsylvania State University.

Host Item Entry  

Dissertations Abstracts International. 86-05B.

Electronic Location and Access  

로그인을 한후 보실 수 있는 자료입니다.

Control Number  

joongbu:657224