Theses
Permanent URI for this collectionhttps://hdl.handle.net/1969.6/1140
Browse
Browsing Theses by Department "Computer Science"
Now showing 1 - 2 of 2
- Results Per Page
- Sort Options
Item Overcoming data limitation challenges in predicting tropical storm surge with interpretable machine learning methods(2023-08) Stanton, Carly; King, Scott; Tissot, Philippe; Wang, WenluThe impacts of climate change have increased the risk of storm surge flooding in coastal areas. Tropical islands are especially vulnerable to the effects of sea level rise and the increase in frequency and intensity of tropical cyclones (TCs). Typically, storm surge prediction is performed using a combination of numerical forecasting models, synoptic forecasting, and statistical methods. Machine learning techniques, particularly convolutional neural networks (CNNs), have shown promise in accurately predicting storm surge levels in the short term. However, deep learning methods are computationally expensive and require large amounts of data to train their models. Often researchers must train neural network models on synthetic data generated by numerical models. The goal of this work is to study the effectiveness of simpler, interpretable models, including random forest (RF) regression, multiple linear regression (MLR), and support vector machine regression (SVR), to predict storm surge in San Juan Bay, Puerto Rico using limited local meteorological and tidal data and hurricane reanalysis data from actual storm events over the last few decades. These algorithms were used to predict surge at five different lead times from one hour to 24 hours and were trained on three different feature sets with two different types of training data windows. Models were trained using a leave-one-out cross-validation (LOOCV) approach, in which data for one TC was separated out for each model as a validation dataset. The performance of the models and different training methods was compared in terms of root mean square error (RMSE), normalized RMSE, and error at peak surge. It was found that an RF model trained on data from only eight TCs was able to predict the peak surge of Hurricane Irma to within about 0.03 m and predicted time of peak surge within three hours at lead times up to 12 hours as long as one extreme TC event, in this case Hurricane Maria, was included in the training data. However, all models failed to accurately predict surge for Hurricane Maria, even when including other high-surge storms in the training data. Other training methods achieved lower RMSE when validated against a peak surge window from the 12 hours prior to 12 hours after peak surge, but could not approach the accuracy of the RF model at predicting the time of peak surge.Item SECURENN: Defeating adversarial neural network attacks with moving target defense and genetic algorithms(2023-05) Romero, Laila Maria; Rubio-Medrano, Carlos; Wang, Wenlu; King, ScottNeural Networks (NNs) have become a critical part of Artificial Intelligence due to their reputation of producing highly accurate outputs with minimal human assistance. NNs are used in various diverse implementations from housing market predictors to medical imaging. Their swift increase in importance and incorporation into our lives have rendered them valuable targets to Adversarial Attacks. Adversarial Attacks are malicious actions aimed to undermine NN model performance, cause misbehavior, and acquire protected information. NNs are used to run many state-of-the-art image classification systems therefore, attacks could be dangerous to the property, health and safety of their users. The most common and successful attacks are gradient based attacks on Image Classification Neural Networks. The defense strategies in existence, such as Adversarial Training, fall short on their ability to protect models against more complex attacks due to their susceptibility to degrade generalization ability in models. This work proposes SecureNN, a defense framework for image classification NNs to increase overall robustness of the models against white-box untargeted Adversarial Attacks. Through the combination of the well- established cybersecurity and Machine Learning techniques of Moving Target Defense, Genetic Algorithm, and Ensemble Learning, SecureNN is able reduce the degraded generalization ability seen in most defense methods as well as minimize the advantages white-box attacks have without incurring in significant cost on the accuracy and speed of the model. SecureNN has been tested extensively on the following four NN architecture types: CNN, ResNet50, Inception, and Inception-ResNet and trained with three common datasets of MNIST, ImageNet and Cifar-10. Each model architecture and dataset were tested against the four highest error rate gradient-based attacks of Fast Gradient Sign Method, Basic Iterative Method, Projected Gradient Descent and Carlini Wagner. The average of 1.5% higher accuracy rates than Adversarial Training and 49.6% higher accuracy rates than Undefended Models exhibited through the experimental phase of our framework substantiates SecureNN’s potential as a defense mechanism effective in increasing NN robustness.