Theses and Dissertations from UMD
Permanent URI for this communityhttp://hdl.handle.net/1903/2
New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a give thesis/dissertation in DRUM
More information is available at Theses and Dissertations at University of Maryland Libraries.
Browse
3 results
Search Results
Item Adversarial Vulnerabilities of Deep Networks and Affordable Robustness(2020) Shafahi, Ali; Goldstein, Thomas A; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Deep learning has improved the performance of many computer vision tasks. However, the features that are learned without extra regularization are not necessarily interpretable. While in terms of generalization, conventionally trained models seem to perform really well, they are susceptible to certain failure modes. Eventhough these catastrophic failure cases rarely happen naturally, an adversary can engineer them by having some knowledge about the design process. Based on the time that the adversary manipulates the system, we can classify threats into evasion attacks or data poisoning attacks. First, we will cover a recently proposed data poisoning threat model that does not assume that the adversary has control over the labeling process. We call this attack the “targeted clean-label” poisoning attack. The proposed attack successfully causes misclassification of a target instance both under end-to-end training and transfer learning scenarios without degrading the overall performance of the classifier on non-target examples. We will then shift our focus to evasion attacks. We will consider two types of inference-time attacks: universal perturbations, and adversarial examples. For universal perturbations, we present an efficient method for perturbation generation. We also propose universal adversarial training for defending against universal perturbations. In the last part of this dissertation, we will present methods for training per-instance robust models in settings where we have limited resources. One case of limited resources is the scarcity of computing power. In this case, we will present our algorithm called “Adversarial Training for Free!” which enables us to train robust models with the same computational cost of conventional/natural training. We achieve this efficiency by simultaneously updating the network parameters and the adversarial perturbation. Another case of limited resources is availability of training data per-class. For this case, we introduce adversarially robust transfer learning.Item Mathematical Model and Framework for Multi-Phase Project Optimization(2016) Shafahi, Ali; Haghani, Ali; Civil Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This research aims to assist investors of “real” tangible assets such as construction projects in making an optimal portfolio of phased and regular projects which will yield the best financial outcome calculated in terms of discounted cash flow of future anticipated revenues and costs. We use optimization techniques to find the optimal timing and phasing of a single project that has the potential of being decomposed into smaller sequential phases. Existence of uncertainties is inevitable especially in cases in which we are planning for long durations. In the presence of these uncertainties, full upfront commitment to large projects may jeopardize the rationality of investments and cause substantial economic risks. Breaking a big project into smaller stages (phases) and implementing a staged development is a potential mechanism to hedge the risk. Under this approach, by adding managerial flexibilities, we may choose to abandon a project at any time once the uncertain outcomes are not favorable. In addition to the benefits resulting from hedging unfavorable risks, phasing a project can transform a financially infeasible project into a feasible one due to less load on capital budgets during each time. Once some phases of a project are delayed and planned to be implemented sequentially, it is important to prepare the infrastructure required for their future development. Initially, we present a Mixed Integer Programming (MIP) model for the deterministic case with no uncertainties that considers interrelationships between phases of projects such as scheduling and costs (economy of scales) in addition to the initial infrastructural investment required for implementation of future phases. Pairing possible phases of a project and doing them in parallel is beneficial due to positive synergies between phases but on the downside requires larger capital investments. Unavailability of enough budgets to fully develop a profitable project will cause the investment to be carried out in different phases e.g. during times when the required capital for developing the next phase (or group of phases) is available. After, presenting the model for the deterministic case, we present a scenario-based multi-stage MIP model for the stochastic case. The source of uncertainty considered is future demand that is modeled using a trinomial lattice. We then present two methods for solving the stochastic problem and finding the value of the here and now decision variable (the size of the infrastructure/foundation). Finding the value of the here and now decision variable for all scenarios using a novel technique that does not require solving all the scenarios is the first method. The second method combines simulation and optimization to find good solutions for the here and now decision variable. Lastly, we present a MIP for the deterministic multi-project case. In this setting, projects could have multiple phases. The MIP will help the managers in making the project selection and scheduling decision simultaneously. It will also assist the managers in making appropriate decisions for the size of the infrastructure and the implementation schedule of the phases of each project. To solve this complex model, we present a pre-processing step that helps reduce the size of the problem and a heuristic that finds good solutions very fast.Item BIDDING ON PROJECTS BASED ON PREVIOUS WORKS AND EMINENCE, A CONTRACTORS' VIEW POINT(2012) Shafahi, Ali; Haghani, Ali; Civil Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Nowadays, a project-winning environment is very competitive. In order to be successful and make profit, a contractor should start planning in advance and decide about what projects to bid. A prominent contractor should bid on projects for which his chances of winning are good enough or on projects for which the profit is high enough such that bidding would be worth consuming the resources needed for preparing the bid. The chances of winning bids are related to many factors but among all, degree of eminence (previous works) and price are the most important. In this thesis we have developed an optimization model that maximizes an index that takes both of these factors into consideration. Genetic Algorithm is used to solve this optimization model.The output of this model is the most beneficial set of projects and their respected optimal bid markups that will help the contractor make the most intuitive selection which, in return, benefits him/her the most at present and in the future.