University of Maryland, Baltimore County
Not a member yet
17643 research outputs found
Sort by
Multi-Source Option-Based Policy Transfer
No full textReinforcement learning algorithms are very effective at learning policies (mappings from states to actions) for specific well defined tasks, thereby allowing an agent to learn how to behave without extensive deliberation. However, if an agent must complete a novel variant of a task that is similar to, but not exactly the same as, a previous version for which it has already learned a policy, learning must begin anew and there is no benefit to having previously learned anything. To address this challenge, I introduce novel approaches for policy transfer. Policy transfer allows the agent to follow the policy of a previously solved, but different, task (called a source task) while it is learning a new task (called a target task). Specifically, I introduce option-based policy transfer (OPT). OPT enables policy transfer by encapsulating the policy for a source task in an option Sutton, Precup, & Singh 1999), which allows the agent to treat the policy of a source task as if it were a primitive action. A significant advantage of this approach is that if there are multiple source tasks, an option can be created for each of them, thereby enabling the agent to transfer knowledge from multiple sources and to combine their knowledge in useful ways. Moreover, this approach allows the agent to learn in which states of the world each source task is most applicable. OPT's approach to constructing and learning with options that represent source tasks allows OPT to greatly outperform existing policy transfer approaches. Additionally, OPT can utilize source tasks that other forms of transfer learning for reinforcement learning cannot. Challenges for policy transfer include identifying sets of source tasks that would be useful for a target task and providing mappings between the state and action spaces of source and target tasks. That is, it may not be useful to transfer from all previously solved source tasks. If a source task has a different state or action space than the target task, then a mapping between these spaces must be provided. To address these challenges, I introduce object-oriented OPT (OO-OPT), which leverages object-oriented MDP (OO-MDP) (Diuk, Cohen, & Littman 2008) state representations to automatically detect related tasks and redundant source tasks, and to provide multiple useful state and action space mappings between tasks. I also introduce methods to adapt value function approximation techniques (which are useful when the state space of a task is very large or continuous) to the unique state representation of OO-MDPs
Indeterminate Place. Site-specific art in the age of global de-territorialization.
No full textThe aim of this research is to investigate, through the lenses of art, themes of contemporary life such as nomadism, de-territorialization, out-of-placeness and de-personalization--that is, some of the most essential aspects that define the post- modern and the post-industrial condition. An art that aspires to engage with topics of presence, place, and movement, has to be participant-oriented and site-oriented in order to reflect and express the real conditions of a body in a specific space. Such components as interaction, participation, comprehensiveness, immersion, and perceptiveness are also required artistic strategies that best assure the purpose of a site-oriented art. This work is also an inquiry into the concept of Indetermination, which is the quality that informs my work overall. Indetermination is not yet an artistic strategy but rather a structural component. It condenses and projects ideas of differentiation, multiplicity, spatiality, ambiguity, openness, and anarchism. This work informs the spatial-temporal qualities of an architectural environment, where the movement of a body in space takes place. The goal of this thesis is that of experimenting and verifying moments of location inside that heterotopic entity (Michel Foucault) and de-located site (Jean Baudrillard) represented by the gallery space, with the purpose of activating a dialectical consid- eration of the production of space (Henri Lefebvre). In other words, through dispersed micro-narratives of rhizomatic territorialization (Gilles Deleuze and F�lix Guattari), through dialectical oppositions demarcating fragments of places and non- places (Miwon Kwon), I aspire to stimulate a reciprocal reflection between myself and an audience upon the spatial and temporal qualities of the sites we inhabit in our lives, in this age of mobility and nomadism
Modeling Individual Nodes in Link Prediction
No full textThe question of how to predict which links will form in a graph, given the graph's history, is an open research problem in computer science. There are many different approaches to the link prediction problem, one of which involves building a set of features for pairs of nodes and using supervised learning to build a model that predicts when these pairs of nodes will link. Typically, this model is learned over the entire graph. In this thesis, I investigate building this model over each individual node in an attempt to learn the particular ways in which that node behaves before making predictions about it. In addition, research into link prediction to date lacks intelligent ways of utilizing the graph over large timespans. To address this, I introduce a variety of ways to include temporality into the link prediction process by introducing new ways of using existing features
The Environmental and Economic Benefits of Stream Restoration: An Application to Stream Restoration in Maryland
No full textThe dissertation analyzes the economic value of improvements resulting from stream restoration within the Chesapeake Bay watershed and reviews what is known about the economic value of the services provided by restoration and the environmental improvements of stream restoration. Although there is uncertainty about the level of improved services restoration provides studies from the different disciplines help to inform the economic analysis that follows by providing evidence on the types of improvements to the services that are expected from a restoration. To assess the economic benefits of restoring streams requires an analysis of how humans value the services from streams. Some of these services accrue to individuals, who might even pay for them directly. Such services might include, for example, erosion and flood control, and their value can be determined using market data or information from private sector projects. However, many of the benefits of stream restoration are public in nature, such as improvements in water or habitat quality in locations downstream from where restoration occurs. These benefits may be substantial and are likely to be external to any market transaction. This study uses contingent valuation (CV) to measure the value of the following environmental services that may follow from stream restoration; improved water quality, habitat improvement, greater habitat diversity, more erosion and flood control, and overall health of the Chesapeake Bay. Results from the analysis of the CV survey suggest that the annual willingness to pay (WTP) for stream restoration projects within Maryland that improves degraded streams to moderately healthy ranges from 114. When extrapolated to the population of the Greater Baltimore Region the total WTP is 118.5 for a moderate stream restoration improvement. Additionally, by using pictures to evaluate the aesthetic preferences of individuals, respondents prefer streams that have open banks with limited riparian vegetation, and streams without erosion. Respondents also prefer streams that are free of trash, but have no clear preference for the shape of the stream
Linked Data for Software Security Concepts and Vulnerability Descriptions
No full textThe Web is typically our first source of information about new software vulnerabilities, exploits and cyber-attacks. Information is found in semi-structured vulnerability databases as well as in text from security bulletins, news reports, cybersecurity blogs and Internet chat rooms. It can be useful to cybersecurity systems if there is a way to recognize and extract relevant information and represent it as easily shared and integrated semantic data. We describe such an automatic framework that generates and publishes a RDF linked data representation of cybersecurity concepts and vulnerability descriptions extracted from the National Vulnerability Database and other text sources. Entities, relations and concepts are represented using custom ontologies for the cybersecurity domain and also mapped to objects in the DBpedia knowledge base, producing a rich resource of machine-understandable linked data. The resulting cybersecurity linked data collection can be used for many purposes, including automating early vulnerability identification, mitigation and prevention efforts
SENTIMENT ANALYSIS ON TWEETS AND THEIR RELATIONSHIP WITH STOCK MARKET TRENDS
No full textWe investigate whether sentiment derived from micro-blogging site Twitter can be used to identify important events (product launch, quarter results etc.) and help to infer the future movement of the stock. We used the volume and key performance index of Apple Company's financial tweets to identify important events and infer the future movement. We present the results of machine learning algorithms (Na?ve Bayes, Maximum Entropy, and SVM) for classifying the sentiment of Apple Company's financial tweets. Statistical analysis using Granger causality test showed that we were able to infer the movement of Apple Company's stock close price in advance
The Socio-ecological System of Vacant Lot Management for Baltimore City Neighborhoods
No full textThis dissertation provides an in-depth investigation, analysis, and critique of natural and human systems involving vacant land in neighborhoods of Southwest Baltimore City, Maryland. Using an interdisciplinary framework of theoretical, conceptual, and methodological approaches from urban ecology, political ecology, and environmental justice, this dissertation challenges popular notions from academia, science, policy, and social justice concerning vacant land in distressed inner-city neighborhoods. By challenging these notions, the dissertation aims to advance intellectual and theoretical perspectives about nature, community, safety, and health of these communities. In order to provide a foundation for understanding the interdisciplinary nature of this dissertation, the dissertation discusses challenges, barriers, and opportunities related to designing and conducting interdisciplinary research. After laying this foundation, the chapters of the dissertation are presented under the theme of Ownership, Responsibility, and Vacant Lot Management. With respect to this theme, the chapters explore topics related to ecological, social, economic, and political systems associated with disinvested communities of Southwest Baltimore City, Maryland. In each of the chapters, meanings for ecology, environment and nature are explored in the context of ownership and responsibility for vacant land. The final chapter of the dissertation provides an interdisciplinary analysis of key findings and conclusions regarding the ecological, social and political landscape of disinvestment and the potential for procedural justice for Southwest Baltimore
Perceptions of Wind Power, Community, and Renewable Energy Landscapes
No full textRenewable energy enjoys broad popularity as an abstract concept, yet specific cases of industrial renewable energy development consistently encounter local opposition commonly characterized by the phrase `not in my backyard,' or NIMBY. Scholars have recently attempted a broader, more thorough understanding of wind opposition focusing on discursive forms of investigation, but even these efforts are limited by unspoken assumptions of motivation and aim. These attempts have led to a sophisticated understanding of local scale arguments mobilized by wind opponents, but have ignored arguments on other scales. In a case study of Keyser, West Virginia, I discovered a broad range of arguments not included in the scholarly literature. This thesis will investigate arguments on scales including the global, national, regional, local, and individual. I will connect the local scale objections to broader scales through the lenses of landscape and placemaking theory, and I will investigate the emergent discourse of health impacts and Wind Turbine Syndrome through embodiment theory and parallel phenomena
Time-Resolved Hyperspectral Imaging of Pulsed-Laser Induced Fluorescence: Food Safety Inspection
No full text`Visual inspection of produce fields for signs of fecal contamination and animal intrusion prior to harvest is a currently prescribed method for reducing risk of foodborne illnesses. This report details the continuing development of a fluorescence imaging system to augment human visual inspection. Previous research established that chlorophyll and related compounds, commonly found in feces and injury sites on plants, can be detected using fluorescent responses to UV excitation. In this study, a time-resolved multispectral imaging system was modified to allow hyperspectral line-scan imaging of fluorescent responses to 355-nm pulsed-laser excitation. Addition of a spectral adapter allowed acquisition of line images representing one spatial dimension with full spectral information for each spatial pixel location. Full-object images can be created by concatenating sequential line-scan images. For line-scan imaging, full-object illumination is inefficient. To better fit the illumination to the line-scan imaging field, laser line-expansion was achieved using a Powell lens. Illumination efficiency was 28.5% using the Powell lens compared to 3.0% using simple optical expansion. To test the modified system, spinach was inoculated with 1:2, 1:10, 1:100, and 1:200 dilutions of bovine manure. Using detection based on visual observation, a 100% detection rate was found for all dilutions. Automated detection rates using estimates of fluorescent decay were 100%, 100%, 100%, and 82 % for the 1:2, 1:10, 1:100, and 1:200 dilutions, respectively. These results suggest this technology has potential for the development of a commercial system for pre-harvest detection of fecal contamination and signs of animal intrusion
Novel Hybrid Chromatofocusing Methods for Protein Purification
No full textThe efforts made to-date to alleviate the downstream challenges faced by the biopharmaceutical industry have been mainly focused on developing novel chromatographic column packings with either higher ligand densities to accommodate larger production capacity requirements or novel ligand groups that exhibit more than one interaction mode to increase selectivity. However, despite this earlier work, there are still unexploited interactions associated with the functional groups present on the column packings that, if optimized, may lead to novel chromatographic separation techniques. Thus, the rational behind the studies investigated here is to provide innovative separation methods based on hybrid chromatofocusing techniques and employ these unexploited interactions, which may be useful in protein purification process development for the biopharmaceutical industry. A comprehensive optimization method capable of exploiting the synergetic effects both the pH and ionic strength on ion-exchange column packing has not yet been developed. Consequently, one primary research objective of this study is to establish the usefulness of employing combined pH and ionic strength gradients to obtain elements of orthogonal two-dimension chromatography in one ion-exchange column that is suitable for the preparative purification of protein in both dilute and non-dilute regimes. Another main objective is to take the concepts developed in this study for ion-exchange chromatography and apply them in affinity and mixed-mode chromatography, where the ligands on the column packings may exhibit electrostatic interactions as well as hydrophobic, hydrogen bonding and/or affinity interactions, and where pH gradients play a major role in the protein adsorption/desorption process. Lastly, this work aims to increase the understanding of the technique of chromatofocusing based on the use of modern theoretical and experimental tools, and to use this understanding to develop novel hybrid chromatofocusing methods. For this purpose, the development of a computer-aided optimization methodology was also performed which allows efficient chromatographic system identification and optimized design. In this way, the computer simulation methods described here go well beyond any previous attempts at simulations in this area