Physical Sciences and Mathematics Commons^™

Challenges In Large-Scale Machine Learning Systems: Security And Correctness, Emad Alsuwat

Theses and Dissertations

In this research, we address the impact of data integrity on machine learning algorithms. We study how an adversary could corrupt Bayesian network structure learning algorithms by inserting contaminated data items. We investigate the resilience of two commonly used Bayesian network structure learning algorithms, namely the PC and LCD algorithms, against data poisoning attacks that aim to corrupt the learned Bayesian network model.

Data poisoning attacks are one of the most important emerging security threats against machine learning systems. These attacks aim to corrupt machine learning models by con- taminating datasets in the training phase. The lack of resilience of …

Integration Of Chlorpyrifos Acetylcholinesterase Inhibition, Water Temperature, And Dissolved Oxygen Concentration Into A Regional Scale Multiple Stressor Risk Assessment Estimating Risk To Chinook Salmon, Wayne G. Landis, Valerie R. Chu, Scarlett E. Graham, Meagan J. Harris, April J. Markiewicz, Chelsea J. Mitchell, Katherine E. Von Stackelberg, John D. Stark

IETC Publications

We estimated the risk to populations of Chinook salmon (Oncorhynchus tshawytscha) due to chlorpyrifos (CH), water temperature (WT), and dissolved oxygen concentration (DO) in 4 watersheds in Washington State, USA. The watersheds included the Nooksack and Skagit Rivers in the Northern Puget Sound, the Cedar River in the Seattle–Tacoma corridor, and the Yakima River, a tributary of the Columbia River. The Bayesian network relative risk model (BN‐RRM) was used to conduct this ecological risk assessment and was modified to contain an acetylcholinesterase (AChE) inhibition pathway parameterized using data from CH toxicity data sets. The completed BN‐RRM estimated risk at a …

Using Bayesian Networks To Predict Risk To Estuary Water Quality And Patterns Of Benthic Environmental Dna In Queensland, Scarlett E. Graham, Anthony A. Chariton, Wayne G. Landis

Institute of Environmental Toxicology & Chemistry Publications

Predictive modeling can inform natural resource management by representing stressor-response pathways in a logical way and quantifying the effects on selected endpoints. This study demonstrates a risk assessment model using the Bayesian network-relative risk model (BNRRM) approach to predict water quality and; for the first time, eukaryote environmental DNA (eDNA) data as a measure of benthic community structure. Environmental DNA sampling is a technique for biodiversity measurements that involves extracting DNA from environmental samples, amplicon sequencing a targeted gene, in this case the 18s rDNA gene which targets eukaryotes, and matching the sequences to organisms. Using a network of probability …

Symptom-Aware Hybrid Fault Diagnosis Algorithm In The Network Virtualization Environment, Yuze Su, Xiangru Meng, Xiaoyang Han, Qiaoyan Kang

Turkish Journal of Electrical Engineering and Computer Sciences

As an important technology in next-generation networks, network virtualization has received more and more attention. Fault diagnosis is the crucial element for fault management and it is the process of inferring the exact failure in the network virtualization environment (NVE) from the set of observed symptoms. Although various traditional fault diagnosis algorithms have been proposed, the virtual network has some new characteristics, which include inaccessible fault information of the substrate network, inaccurate network observations, and a dynamic embedding relationship. To solve these challenges, a symptom-aware hybrid fault diagnosis (SAHFD) algorithm in the NVE is proposed in this paper. First, a …