Mountain Scholar :: Browsing by Author "Anderson, Charles W., committee member"

Browsing by Author "Anderson, Charles W., committee member"

Now showing 1 - 20 of 21

Open Access
Assessment of protein-protein interfaces using graph neural networks
(Colorado State University. Libraries, 2021) Virupaksha, Yashwanth Reddy, author; Ben-Hur, Asa, advisor; Anderson, Charles W., committee member; Adams, Henry Hugh, committee member
Proteins are fundamental building blocks of cellular function. They systematically interact with other proteins to make life happen. Understanding these protein-protein interactions is important for obtaining a detailed understanding of protein function and to enable the process of drug and vaccine design. Experimental methods for studying protein interfaces including X-ray crystallography, NMR, and Cryo-electron microscopy, are expensive, time consuming, and sometimes unsuccessful due to the unstable nature of many protein-protein interactions. Computational docking experiments are a cheap and fast alternative. Docking algorithms produce a large number of potential solutions that are then ranked by quality. However, current scoring methods are not good enough for finding a docking solution that is close to the native structure. That has led to the development of machine learning methods for this task. These methods typically involve extensive engineering of features to describe the protein complex, and are not very successful at identifying good quality solutions among the top ranks. In this thesis, we propose a scoring technique that uses graph neural networks that function at the atomic level to learn the interfaces of docked proteins without the need for feature engineering. We evaluate our model and show that it performs better than commonly used docking methods and deep learning methods that use 3D CNNs.
Open Access
COVID-19 misinformation on Twitter: the role of deceptive support
(Colorado State University. Libraries, 2022) Hashemi Chaleshtori, Fateme, author; Ray, Indrakshi, advisor; Anderson, Charles W., committee member; Malaiya, Yashwant K., committee member; Adams, Henry, committee member
Social media platforms like Twitter are a major dissemination point for information and the COVID-19 pandemic is no exception. But not all of the information comes from reliable sources, which raises doubts about their validity. In social media posts, writers reference news articles to gain credibility by leveraging the trust readers have in reputable news outlets. However, there is not always a positive correlation between the cited article and the social media posting. Targeting the Twitter platform, this study presents a novel pipeline to determine whether a Tweet is indeed supported by the news article it refers to. The approach follows two general objectives: to develop a model capable of detecting Tweets containing claims that are worthy of fact-checking and then, to assess whether the claims made in a given Tweet are supported by the news article it cites. In the event that a Tweet is found to be trustworthy, we extract its claim via a sequence labeling approach. In doing so, we seek to reduce the noise and highlight the informative parts of a Tweet. Instead of detecting erroneous and invalid information by analyzing the propagation patterns or ensuing examination of Tweets against already proven statements, this study aims to identify reliable support (or lack thereof) before misinformation spreads. Our research reveals that 14.5% of the Tweets are not factual and therefore not worth checking. An effective filter like this is especially useful when looking at a platform such as Twitter, where hundreds of thousands of posts are created every day. Further, our analysis indicates that among the Tweets which refer to a news article as evidence of a factual claim, at least 1% of those Tweets are not substantiated by the article, and therefore mislead the reader.
Open Access
Deep transfer learning for prediction of health risk behaviors in adolescent psychiatric patients
(Colorado State University. Libraries, 2021) Kentopp, Shane, author; Conner, Bradley T., advisor; Prince, Mark A., committee member; Henry, Kimberly L., committee member; Anderson, Charles W., committee member
Binge drinking and non-suicidal self-injury are significant health-risk behaviors that are often initiated during adolescence and contribute to a host of negative outcomes later in life. Selective prevention strategies are targeted toward individuals most at-risk for developing these behaviors. Traditionally, selective interventions are tailored based on risk factors identified by human experts. Machine learning algorithms, such as deep neural networks, may improve the effectiveness of selective interventions by accounting for complex interactions between large numbers of predictor variables. However, their use in psychological research is limited due to the tendency to overfit and the need for large volumes of training data. Deep transfer learning can overcome this limitation by leveraging samples of convenience to facilitate training deep neural networks in small, clinically relevant samples. The author trained deep neural networks on data from a sample of adolescent psychiatric inpatients to retrospectively classify individuals according to their history of alcohol misuse and nonsuicidal self-injury. Next, the performance of these models was compared to deep neural networks that were pretrained in a convenience sample of college undergraduates and fine-tuned in the sample of psychiatric patients. Deep transfer learning did not improve classification accuracy but buffered against overfitting. The deep neural networks that were not pretrained maintained maximum classification accuracy for a very small number of training epochs before performance deteriorated due to overfitting the training data. Conversely, the pretrained networks maintained their maximum classification accuracy across many training epochs and performance was not hindered by overfitting. This suggests that convenience samples can be utilized to reduce the risk of overfitting when training complex deep neural networks on small clinical samples. In the future, this process may be employed to facilitate powerful predictive models that inform selective prevention programs and contribute to the reduction of health risk behavior prevalence amongst vulnerable adolescent populations.
Open Access
Design and evaluation of the FAMILIAR tool
(Colorado State University. Libraries, 2014) Jaksic, Aleksandar, author; France, Robert B., advisor; Anderson, Charles W., committee member; Ghosh, Sudipto, committee member; Troup, Lucy J., committee member
Software Product Line Engineering (SPLE) aims to efficiently produce multiple software products, on a large scale, that share a common set of core development features. Feature Modeling is a popular SPLE technique used to describe variability in a product family. FAMILIAR (FeAture Model scrIpt Language for manipulation and Automatic Reasoning) is a Domain-Specific Modeling Language (DSML) for manipulating Feature Models (FMs). One of the strengths of the FAMILIAR language is that it provides rich semantics for FM composition operators (aggregate, merge, insert) as well as decomposition operators (slice). The main contribution of this thesis is to provide an integrated graphical modeling environment that significantly improves upon the initial FAMILIAR framework that was text-based and consisted of loosely coupled parts. As part of this thesis we designed and implemented a new FAMILIAR Tool that provides (1) a fast rendering framework for the graphically representing feature models, (2) a configuration editor and (3) persistence of feature models. Furthermore, we evaluated the usability of our new FAMILIAR Tool by performing a small experiment primarily focusing on assessing quality aspects of newly authored FMs as well as user effectiveness and efficiency.
Open Access
Extending and validating the stencil processing unit
(Colorado State University. Libraries, 2016) Rajasree, Revathy, author; Rajopadhye, Sanjay, advisor; Pasricha, Sudeep, committee member; Anderson, Charles W., committee member
Stencils are an important class of programs that appear in the core of many scientific and general-purpose applications. These compute-intensive kernels can benefit heavily from the massive compute power of accelerators like the GPGPU. However, due to the absence of any form of on-chip communication between the coarse-grain processors on a GPU, any data transfer/synchronization between the dependent tiles in stencil computations has to happen through the off-chip (global) memory, which is quite energy-expensive. In the road to exascale computing, energy is becoming an important cost metric. The need for hardware and software that can collaboratively work towards reducing energy consumption of a system is becoming more and more important. To make the execution of dense stencils more energy efficient, Rajopadhye et al. proposed the GPGPU-based accelerator called Stencil Processing Unit that introduces a simple neighbor-to-neighbor communication between the Streaming Multiprocessors (SM) on the GPU, thereby allowing some restricted data sharing between consecutive threadblocks. The SPU includes special storage units, called Communication Buffers, to orchestrate this data transfer and also provides an explicit mechanism for inter-threadblock synchronization by way of a special instruction. It claims to achieve energy-efficiency, compared to GPUs, by reducing the number of off-chip accesses in stencils which in turn reduces the dynamic energy overhead. Uguen developed a cycle-accurate performance simulator for the SPU, called SPU-Sim, and evaluated it using a matrix multiplication kernel which was not suitable for this accelerator. This work focuses on extending the SPU-Sim and evaluating the SPU architecture using a more insightful benchmark. We introduce a producer-consumer based inter-block synchronization approach on the SPU, which is more efficient than the previous global synchronization, and an overlapped multi-pass execution model in the SPU runtime system. These optimizations have been implemented into SPU-Sim. Furthermore, the existing GPUWattch power model in the simulator has been refined to provide better power estimates for the SPU architecture. The improved architecture has been evaluated using a simple 2-D stencil benchmark and we observe an average of 16% savings in dynamic energy on SPU compared to a fairly close GPU platform. Nonetheless, the total energy consumption on SPU is still comparatively high due to the static energy component. This high static energy on SPU is a direct impact of the increased leakage power of the platform resulting from the inclusion of special load/store units. Our conservative estimates indicate that replacing the current design of these L/S units with DMA engines can bring about a 15% decrease in the current leakage power of the SPU and this can help SPU outperform GPU in terms of energy.
Open Access
Finding a solution for the tradeoff between time, cost and sustainability/LEED credits for new construction
(Colorado State University. Libraries, 2018) Kumar, Shantanu, author; Mehany, Mohammed S. Hashem M., advisor; Guggemos, Angela Acree, committee member; Anderson, Charles W., committee member
Project complexity generated tradeoffs in construction, which evolved over decades. This research focuses on the tradeoff between time-cost and sustainability represented in the LEED credits (Materials and Resources in particular). The research was broken down into preliminary and validation studies, wherein the preliminary study used an exhaustive search to find the optimized solution. In validation case study, the size of dataset increased exponentially, and it became computationally incompatible to find the optimized solution. Genetic Algorithm (GA) was hence used to find the optimized solution based on priority factors entered by the user. Usage of GA was validated using the preliminary study data and then applied to the validation study data. A tradeoff could be seen between the priority factors and the optimized solution. It was found that the optimization model was successful in minimizing the time and cost, concurrently maximizing the credits for a validation case study conducted for a real-life project.
Open Access
Improving gesture recognition through spatial focus of attention
(Colorado State University. Libraries, 2018) Narayana, Pradyumna, author; Draper, Bruce A., advisor; Beveridge, Ross J., committee member; Anderson, Charles W., committee member; Peterson, Christopher, committee member
Gestures are a common form of human communication and important for human computer interfaces (HCI). Most recent approaches to gesture recognition use deep learning within multi- channel architectures. We show that when spatial attention is focused on the hands, gesture recognition improves significantly, particularly when the channels are fused using a sparse network. We propose an architecture (FOANet) that divides processing among four modalities (RGB, depth, RGB flow, and depth flow), and three spatial focus of attention regions (global, left hand, and right hand). The resulting 12 channels are fused using sparse networks. This architecture improves performance on the ChaLearn IsoGD dataset from a previous best of 67.71% to 82.07%, and on the NVIDIA dynamic hand gesture dataset from 83.8% to 91.28%. We extend FOANet to perform gesture recognition on continuous streams of data. We show that the best temporal fusion strategies for multi-channel networks depends on the modality (RGB vs depth vs flow field) and target (global vs left hand vs right hand) of the channel. The extended architecture achieves optimum performance using Gaussian Pooling for global channels, LSTMs for focused (left hand or right hand) flow field channels, and late Pooling for focused RGB and depth channels. The resulting system achieves a mean Jaccard Index of 0.7740 compared to the previous best result of 0.6103 on the ChaLearn ConGD dataset without first pre-segmenting the videos into single gesture clips. Human vision has α and β channels for processing different modalities in addition to spatial attention similar to FOANet. However, unlike FOANet, attention is not implemented through separate neural channels. Instead, attention is implemented through top-down excitation of neurons corresponding to specific spatial locations within the α and β channels. Motivated by the covert attention in human vision, we propose a new architecture called CANet (Covert Attention Net), that merges spatial attention channels while preserving the concept of attention. The focus layers of CANet allows it to focus attention on hands without having dedicated attention channels. CANet outperforms FOANet by achieving an accuracy of 84.79% on ChaLearn IsoGD dataset while being efficient (≈35% of FOANet parameters and ≈70% of FOANet operations). In addition to producing state-of-the-art results on multiple gesture recognition datasets, this thesis also tries to understand the behavior of multi-channel networks (a la FOANet). Multi- channel architectures are becoming increasingly common, setting the state of the art for performance in gesture recognition and other domains. Unfortunately, we lack a clear explanation of why multi-channel architectures outperform single channel ones. This thesis considers two hypotheses. The Bagging hypothesis says that multi-channel architectures succeed because they average the result of multiple unbiased weak estimators in the form of different channels. The Society of Experts (SoE) hypothesis suggests that multi-channel architectures succeed because the channels differentiate themselves, developing expertise with regard to different aspects of the data. Fusion layers then get to combine complementary information. This thesis presents two sets of experiments to distinguish between these hypotheses and both sets of experiments support the SoE hypothesis, suggesting multi-channel architectures succeed because their channels become specialized. Finally we demonstrate the practical impact of the gesture recognition techniques discussed in this thesis in the context of a sophisticated human computer interaction system. We developed a prototype system with a limited form of peer-to-peer communication in the context of blocks world. The prototype allows the users to communicate with the avatar using gestures and speech and make the avatar build virtual block structures.
Open Access
Information-augmented building information models (BIM) to inform facilities management (FM) guidelines
(Colorado State University. Libraries, 2019) Sadeghi, Marjan, author; Grigg, Neil, advisor; Elliot, Jonathan W., advisor; Mehany, Mohammed S. Hashem M., committee member; Anderson, Charles W., committee member
The asset portfolios of Higher Education Institutions (HEI) typically incorporate a highly diverse collection of buildings with various and often shared campus uses. These facilities are typically at different points in their operational lifecycle, have different characteristics, rehabilitation cost, maintenance costs, and mission criticality. In the resource-constrained context of higher education Facilities Management (FM), building data for all facilities needs to be integrated within a highly-informed decision-making process to promote efficient operation. Further, efficient building FM workflows depend upon accurate, reliable, and timely information for various building-specific systems, components, and elements. Traditional Facilities Information Management (FIM) platforms and processes have been shown to be inefficient and limited for capturing and delivering the extensive and comprehensive data needed for FM decision making. Such inefficiencies include, but are not limited to, information loss, inconsistencies of the available data, and manual data re-entry at construction-to-operation handover and project close out. Building Information Models (BIMs) are capable of integrating large quantities of data and have been recognized as a compelling tool for facility life-cycle information management. BIMs provide an object-oriented, parametric, 3D environment where meaningful objects with intelligent behavior can contain geometric and non-geometric data. This capability makes BIMs a powerful tool for use beyond building visualization. Furthermore, BIM authoring tools are capable of automatically integrating data with FM technologies. Although BIMs have the potential to provide a compelling tool to capture, deliver, validate, retrieve, exchange, and analyze facility lifecycle information, implementation of BIMs for FM handover and integration within the context of FIM remains limited. A plethora of academic and industry efforts strive to address various aspects of BIM interoperability for handing over building data for implementation in post-construction building operation workflows. Attempts to incorporate BIMs in FIM have generally focused on one of five domains; what information is to be exchanged, how, when, by whom, and why. This three-manuscript dissertation explores FM handover information exchange scenarios and provides a comprehensive, object-oriented BIM solution that identifies the requirements for model content for FM- specific needs. The results formalize an appropriate process and structured framework to deliver BIM content using FM-specific terminologies and taxonomies. BIMs created for design and construction using this framework provide a suitable 3D resource for post-handover FM and building operation. The BIM development framework presented herein can facilitate automated model data validation at project close out and the exchange of AEC data with FIM systems. This modeling process can reduce the need for manual data re-entry or interpretation by FM stakeholders during building operation. This study defines requirements for model Exchange Objects (EOs) to meet FM data Exchange Requirements (ERs) in conjunction with the established Industry Foundation Classes (IFC). The ERs, retrieved from closeout deliverables, are mapped to appropriate IFC Model View Definition (MVD) concepts for EOs, which ultimately provide the technical solution for the FM handover exchange scenario. These concepts determine required entities, their relationships, and properties. The author further translated the concepts to the ERs of Level of Development (LOD) definitions to provide the means for an owner to formalize conveyance of geometric requirements. To formalize a BIMs semantic requirements, not addressed in the LOD schema, this study introduces Level of Semantics (LOS) by mapping ERs to IFC categories and their respective property definitions. The results also include development of an implementation agreement, which customizes the FM handover IFC Model View (MV) according to an organization's specific needs. The IFC MV implementation agreement establishes a common understanding of the FM handover MV content in alliance with the buildingSMART Data Dictionary (bsDD) schema. The modularized and repeatable nature of the resulting framework facilitates its implementation to convey FIM data exchange requirements for future projects.
Open Access
Investigation of a nonlinear controller that combines steady state predictions with integral action
(Colorado State University. Libraries, 2010) Hodgson, David A., author; Duff, William S., advisor; Young, Peter M., advisor; Olsen, Daniel B., committee member; Anderson, Charles W., committee member
Cross-flow water-to-air heat exchangers are a common element in heating ventilating and air conditioning (HVAC) systems. In a typical configuration the outlet air temperature is controlled by the flow rate of water through the coil. In this configuration the heat exchanger exhibits non-linear dynamics. In particular the system has variable gain. Variable gain presents a challenge for the linear controllers that are typically used to control the outlet air temperature. To ensure stability over the entire operating range controllers need to be tuned at the highest gain state. This leads to sluggish response in lower gain states. Previous research has shown the use of steady state predictions of the flow rate needed to produce zero steady state error has improved the transient response of a heat exchanger. In this project a nonlinear controller that provides smooth mixing between steady state predictions and integral control was introduced. Bounds for the steady state error introduced by the controller were theoretically derived and experimentally verified. The controller outperformed a properly tuned nominal PI controller for both input tracking and disturbance rejection.
Open Access
Machine learning methods to facilitate optimal water allocation and management in irrigated river basins to comply with water law
(Colorado State University. Libraries, 2019) Rohmat, Faizal Immaddudin Wira, author; Labadie, John W., advisor; Gates, Timothy K., advisor; Bailey, Ryan T., committee member; Anderson, Charles W., committee member
The sustainability issues facing irrigated river basins are intensified by legal and institutional regulations imposed on the hydrologic system. Although solutions that would boost water savings and quality might prove to be feasible, such imposed institutional constraints could veto their implementation, rendering them legally ineffectual. The problems of basin-scale irrigation water management in a legally-constrained system are exemplified in the central alluvial valley of the Lower Arkansas River Basin (LARB) in Colorado, USA, and in the Tripa River Basin in Indonesia. In the LARB, water and land best management practices (BMPs) have been proposed to enhance the environment, conserve water, and boost productivity; however, the legal feasibility of their implementation in the basin hinder BMP adoption. In the Tripa river basin, the rapid growth of water demand pushes the proposal of new reservoir construction. However, inadequate water availability and the lack of water law enforcement requires the basin to seek water from adjacent basins, thereby raising legal and economic feasibility issues. To address these issues, an updated version of a decision support system (DSS) named River GeoDSS has been employed to model basin-scale behavior of the LARB for both historical (baseline) and BMP implementation scenarios. River GeoDSS uses GeoMODSIM as its water allocation component, which also handles water rights and uses a deep neural network (DNN) functionality to emulate calibrated regional MODFLOW-SFR2 models in modeling complex stream-aquifer interactions. The use of DNNs for emulation if critical for extrapolating the results of MODFLOW-SFR2 simulations to un-modeled portions of the basin and for compute-efficient analysis. The BMP implementations are found to introduce significant alterations to streamflows in the LARB, including shortages in flow deliveries to water right demands and in flow deficits at the Colorado-Kansas Stateline. To address this, an advanced Fuzzy-Mutation Linear Particle Swarm Optimization (Fuzzy-MLPSO) metaheuristic algorithm is applied to determine optimal operational policies for a new storage account in John Martin Reservoir for use in mitigating the side-effects of BMP implementation on water rights and the interstate compact. Prior to the implementation of Fuzzy-MLPSO, a dedicated study is conducted to develop the integration between MLPSO and GeoMODSIM, where it is applied in addressing the water allocation issue in the Tripa River Basin. The coupling of simulation (GeoMODSIM) and optimization (MLPSO) models provides optimal sizing of reservoirs and transbasin diversions along with optimal operation policies. Aside from that, this study shows that MLPSO converges faster compared to the original PSO with sufficiently smaller swarm size. The implementations of Fuzzy-MLPSO in the LARB provided optimal operational rules for a new storage account in John Martin Reservoir dedicated to abating the undesirable impacts of BMP implementation on water rights and Stateline flows. The Fuzzy-MLPSO processes inflow, storage, seasonal, and hydrologic states into divert-to-storage/release-from-storage decisions for the new storage account. Results show that concerns over shortages in meeting water rights demands and deficits to required Stateline flow due to otherwise beneficial BMP implementations can be addressed with optimized reservoir operations.
Open Access
Many hats: intra-trial and reward-level dependent bold activity in the striatum and premotor cortex
(Colorado State University. Libraries, 2012) Peterson, Erik J., author; Seger, Carol A., advisor; Troup, Lucy J., committee member; Anderson, Charles W., committee member
Lesion, drug, single-cell recording, as well as human fMRI studies, suggest dopaminergic projections from VTA/SNc (ventral tagmental area/substantia nigra pars compacta) and cortically driven striatal activity plays a key role in associating sensory events with rewarding actions both by facilitating reward processing and prediction (i.e. reinforcement learning) and biasing and later updating action selection. We, for the first time, isolated BOLD signal changes for stimulus, pre-response, response and feedback delivery at three reward levels. This design allowed us to estimate the degree of involvement of individual striatal regions across these trial components, the reward sensitivity of each component and allowed for a novel comparison of potential (and potentially competing) reinforcement learning computations. Striatal and lateral premotor cortex regions of interest (ROIs) significant activations were universally observed (excepting the ventral striatum) during stimulus presentation, pre-response, response and feedback delivery, confirming these areas importance in all aspects of visuomotor learning. The head of the caudate showed a precipitous drop in activity pre-response, while in the body of the caudate showed no significant changes in activity. The putamen peaked in activity during response. Activation in the lateral premotor cortex was strongest during stimulus presentation, but the drop off was followed by a trend of increasing activity as feedback approached. Both the head and body of the caudate as well as the putamen displayed reward-level sensitivity only during stimulus, while the ventral striatum showed reward sensitivity at both stimulus and feedback. The lack of reward sensitivity surrounding response is inconsistent with theories that the head and ventral striatum encode the value of actions. Which of the three examined reinforcement learning models correlated best with BOLD signal changes varied as a function of trial component and ROI suggesting these regions computations vary depending on task demand.
Open Access
Methodologies for transforming data to information and advancing the understanding of water resources systems towards integrated water resources management
(Colorado State University. Libraries, 2017) Oikonomou, Panagiotis D., author; Fontane, Darrell G., advisor; Waskom, Reagan M., advisor; Grigg, Neil S., committee member; Karavitis, Christos A., committee member; Anderson, Charles W., committee member
The majority of river basins in the world, have undergone a great deal of transformations in terms of infrastructure and water management practices in order to meet increasing water needs due to population growth and socio-economic development. Surface water and groundwater systems are interwoven with environmental and socio-economic ones. The systems' dynamic nature, their complex interlinkages and interdependencies are inducing challenges for integrated water resources management. Informed decision-making process in water resources is deriving from a systematic analysis of the available data with the utilization of tools and models, by examining viable alternatives and their associated tradeoffs under the prism of a set of prudent priorities and expert knowledge. In an era of increasing volume and variety of data about natural and anthropogenic systems, opportunities arise for further enhancing data integration in problem-solving approaches and thus support decision-making for water resources planning and management. Although there is a plethora of variables monitored in various spatial and temporal scales, particularly in the United States, in real life, for water resources applications there are rarely, if ever, perfect data. Developing more systematic procedures to integrate the available data and harness their full potential of generating information, will improve the understanding of water resources systems and assist at the same time integrated water resources management efforts. The overarching objective of this study is to develop tools and approaches to overcome data obstacles in water resources management. This required the development of methodologies that utilize a wide range of water and environmental datasets in order to transform them into reliable and valuable information, which would address unanswered questions about water systems and water management practices, contributing to implementable efforts of integrated water resources management. More specifically, the objectives of this research are targeted in three complementary topics: drought, water demand, and groundwater supply. In this regard, their unified thread is the common quest for integrated river basin management (IRBM) under changing water resources conditions. All proposed methodologies have a common area of application namely the South Platte basin, located within Colorado. The area is characterized by limited water resources with frequent drought intervals. A system's vulnerability to drought due to the different manifestations of the phenomenon (meteorological, agricultural, hydrological, socio-economic and ecological) and the plethora of factors affecting it (precipitation patterns, the supply and demand trends, the socioeconomic background etc.) necessitates an integrated approach for delineating its magnitude and spatiotemporal extent and impacts. Thus, the first objective was to develop an implementable drought management policy tool based on the standardized drought vulnerability index framework and expanding it in order to capture more of drought's multifaceted effects. This study illustrated the advantages of a more transparent data rigorous methodology, which minimizes the need for qualitative information replacing it with a more quantitative one. It is believed that such approach may convey drought information to decision makers in a holistic manner and at the same time avoid the existing practices of broken linkages and fragmentation of reported drought impacts. Secondly, a multi-scale (well, HUC-12, and county level) comparative analysis framework was developed to identify the characteristics of the emergent water demand for unconventional oil and gas development. This effort revealed the importance of local conditions in well development patterns that influence water demand, the magnitude of water consumption in local scales in comparison to other water uses, the strategies of handling flowback water, and the need for additional data, and improved data collection methods for a detailed water life-cycle analysis including the associated tradeoffs. Finally, a novel, easy to implement, and computationally low cost methodology was developed for filling gaps in groundwater level time series. The proposed framework consists of four main components, namely: groundwater level time series; data (groundwater level, recharge and pumping) from a regional physically-based groundwater flow model; autoregressive integrated moving average with external inputs modeling; and the Ensemble Smoother (ES) technique. The methodology's efficacy to predict accurately groundwater levels was tested by conducting three numerical experiments at eighteen alluvial wells. The results suggest that the framework could serve as a valuable tool in gaining further insight of alluvium aquifer dynamics by filling missing groundwater level data in an intermittent or continuous (with relative short span) fashion. Overall, it is believed that this research has important implications in water resources decision making by developing implementable frameworks which advance further the understanding of water systems and may aid in integrated river basin management efforts.
Open Access
Optimal reservoir operations for riverine water quality improvement: a reinforcement learning strategy
(Colorado State University. Libraries, 2011) Rieker, Jeffrey Donald, author; Labadie, John W., advisor; Fontane, Darrell G., committee member; Frevert, Donald K., committee member; Anderson, Charles W., committee member
Complex water resources systems often involve a wide variety of competing objectives and purposes, including the improvement of water quality downstream of reservoirs. An increased focus on downstream water quality considerations in the operating strategies for reservoirs has given impetus to the need for tools to assist water resource managers in developing strategies for release of water for downstream water quality improvement, while considering other important project purposes. This study applies an artificial intelligence methodology known as reinforcement learning to the operation of reservoir systems for water quality enhancement through augmentation of instream flow. Reinforcement learning is a methodology that employs the concepts of agent control and evaluative feedback to develop improved reservoir operating strategies through direct interaction with a simulated river and reservoir environment driven by stochastic hydrology. Reinforcement learning methods have advantages over other more traditional stochastic optimization methods through implicit learning of the underlying stochastic structure through interaction with the simulated environment, rather than requiring a priori specification of probabilistic models. Reinforcement learning can also be coupled with various computing efficiency techniques as well as other machine learning methods such as artificial neural networks to mitigate the "curse of dimensionality" that is common to many optimization methodologies for solving sequential decision problems. A generalized mechanism is developed, tested, and evaluated for providing near-real time operational support to suggest releases of water from upstream reservoirs to improve water quality within a river using releases specifically designated for that purpose. The algorithm is designed to address a variable number of water quality constituents, with additional flexibility for adding new water quality requirements and learning updated operating strategies in a non-stationary environment. The generalized reinforcement learning algorithm is applied to the Truckee River in California and Nevada as a case study, where the federal and local governments are purchasing water rights for the purpose of augmenting Truckee River flows to improve water quality. Water associated with those acquired rights can be stored in upstream reservoirs on the Truckee River until needed for prevention of water quality standard violations in the lower reaches of the river. This study shows that in order for the water acquired for flow augmentation to be fully utilized as a part of a longer-term strategy for water quality management, increased flexibility is required as to how those waters are stored and how well the storage is protected from displacement through reservoir spill during times of high runoff. The results show that with those flexibilities, the reinforcement learning mechanism has the ability to produce both short-term and long-term strategies for the use of the water, with the long-term strategies capable of significantly improving water quality during times of drought over current and historic operating practices. The study also evaluates a number of variations and options for the application of reinforcement learning methods, as well as use of artificial neural networks for function generalization and approximation.
Open Access
Perfect tracking for non-minimum phase systems with applications to biofuels from microalgae
(Colorado State University. Libraries, 2010) Buehner, Michael R., author; Young, Peter M., advisor; Chong, Edwin Kah Pin, committee member; Scharf, Louis L., committee member; Anderson, Charles W., committee member
In a causal setting, a closed-loop control system receives reference inputs (with no a priori knowledge) that it must track. For this setting, controllers are designed that provide both stability and performance (e.g., to meet tracking and disturbance rejection requirements). Often, feedback controllers are designed to satisfy weighted optimization criteria (e.g., weighted tracking error) that are later validated using test signals such as step responses and frequency sweeps. Feedforward controllers may be used to improve the response to measurable external disturbances (e.g., reference inputs). In this way, they can improve the closed-loop response; however, these approaches do not directly specify the closed-loop response. Two controller architectures are developed that allow for directly designing the nominal closed-loop response of non-minimum phase systems. These architectures classify both the signals that may be perfectly tracked by a non-minimum phase plant and the control signals that provide this perfect tracking. For these architectures, perfect tracking means that the feedback error is zero (for all time) in the nominal case (i.e., the plant model is exact) when there are no external disturbances. For the controllers presented here, parts of the feedforward controllers are based on the plant model, while a separate piece is designed to provide the desired level of performance. One of the potential limitations to these designs is that the actual performance will depend upon the quality of the model used. Robustness tools are developed that may be used to determine the expected performance for a given level of model uncertainty. These robustness tools may also be used to design the piece of the feedforward controller that provides performance. There is a tradeoff between model uncertainty and achievable performance. In general, more model uncertainty will result in less achievable performance. Another way to approach the issue of performance is to consider that a good model must either be known a priori or learned via adaptation. In the cases where a good model is difficult to determine a priori, adaptation may be used to improve the models in the feedforward controllers, which will, in turn, improve the performance of the overall control system. We show how adaptive feedforward architectures can improve performance for systems where the model is of limited accuracy. An example application of growing microalgae for biofuel production is presented. Microalgae have the potential to produce enough biofuels to meet the current US fuel demands; however, progress has been limited (in some part) due to a lack of appropriate models and controllers. In the work presented here, models are developed that may be used to monitor the productivity of microalgae inside a photobioreactor and to develop control algorithms. We use experimental data from a functional prototype photobioreactor to validate these models and to demonstrate the advantages of the advanced controller architectures developed here.
Open Access
Predictive energy management strategies for hybrid electric vehicles applied during acceleration events
(Colorado State University. Libraries, 2019) Trinko, David A., author; Bradley, Thomas H., advisor; Quinn, Jason C., committee member; Anderson, Charles W., committee member
The emergence and widespread adoption of vehicles with hybrid powertrains and onboard computing capabilities have improved the feasibility of utilizing predictions of vehicle state to enable optimal energy management strategies (EMS) to improve fuel economy. Real-world implementation of optimal EMS remains challenging in part because of limits on prediction accuracy and computation speed. However, if a finite set of EMS can be pre-derived offline, instead of onboard the vehicle in real time, fuel economy improvements may be possible using hardware that is common in current production vehicles. Acceleration events (AE) are attractive targets for this kind of EMS application due to their high energy cost, probability of recurrence, and limited variability. This research aims to understand how a finite set of EMS might be derived and applied to AEs based on predictions of basic AE attributes to achieve reliable fuel economy improvements. Models of the 2010 Toyota Prius are used to simulate fuel economy for a variety of control strategies, including baseline control, optimal EMS control derived via dynamic programming, and pre-derived control applied with approximate prediction to AEs. Statistical methods are used to identify correlations between AE attributes, optimal powertrain control, and fuel economy results. Then, key AE attributes are used to define AE categorization schemes of various resolutions, in which one pre-derived EMS is applied to every AE in a category. Last, the control strategies are simulated during a variety of drive cycles to predict real-world fuel economy results. By simulating fuel economy improvement for AEs both in isolation and in the context of drive cycles, it was concluded that applying pre-derived EMS to AEs based on predictions of initial and final velocity is likely to enable reliable fuel economy benefits in low-aggression driving.
Open Access
Quantifying the security risk of discovering and exploiting software vulnerabilities
(Colorado State University. Libraries, 2016) Mussa, Awad A. Younis, author; Malaiya, Yashwant, advisor; Ray, Indrajit, committee member; Anderson, Charles W., committee member; Vijayasarathy, Leo, committee member
Most of the attacks on computer systems and networks are enabled by vulnerabilities in a software. Assessing the security risk associated with those vulnerabilities is important. Risk models such as the Common Vulnerability Scoring System (CVSS), Open Web Application Security Project (OWASP) and Common Weakness Scoring System (CWSS) have been used to qualitatively assess the security risk presented by a vulnerability. CVSS metrics are the de facto standard and its metrics need to be independently evaluated. In this dissertation, we propose using a quantitative approach that uses an actual data, mathematical and statistical modeling, data analysis, and measurement. We have introduced a novel vulnerability discovery model, Folded model, that estimates the risk of vulnerability discovery based on the number of residual vulnerabilities in a given software. In addition to estimating the risk of vulnerabilities discovery of a whole system, this dissertation has furthermore introduced a novel metrics termed time to vulnerability discovery to assess the risk of an individual vulnerability discovery. We also have proposed a novel vulnerability exploitability risk measure termed Structural Severity. It is based on software properties, namely attack entry points, vulnerability location, the presence of the dangerous system calls, and reachability analysis. In addition to measurement, this dissertation has also proposed predicting vulnerability exploitability risk using internal software metrics. We have also proposed two approaches for evaluating CVSS Base metrics. Using the availability of exploits, we first have evaluated the performance of the CVSS Exploitability factor and have compared its performance to Microsoft (MS) rating system. The results showed that exploitability metrics of CVSS and MS have a high false positive rate. This finding has motivated us to conduct further investigation. To that end, we have introduced vulnerability reward programs (VRPs) as a novel ground truth to evaluate the CVSS Base scores. The results show that the notable lack of exploits for high severity vulnerabilities may be the result of prioritized fixing of vulnerabilities.
Open Access
Same data, same features: modern ImageNet-trained convolutional neural networks learn the same thing
(Colorado State University. Libraries, 2020) McNeely-White, David G., author; Beveridge, J. Ross, advisor; Anderson, Charles W., committee member; Seger, Carol A., committee member
Deep convolutional neural networks (CNNs) are the dominant technology in computer vision today. Much of the recent computer vision literature can be thought of as a competition to find the best architecture for vision within the deep convolutional framework. Despite all the effort invested in developing sophisticated convolutional architectures, however, it's not clear how different from each other the best CNNs really are. This thesis measures the similarity between ten well-known CNNs, in terms of the properties they extract from images. I find that the properties extracted by each of the ten networks are very similar to each other, in the sense that any of their features can be well approximated by an affine transformation of the features of any of the other nine. In particular, there is evidence that each network extracts mostly the same information as each other network, though some do it more robustly. The similarity between each of these CNNs is surprising. Convolutional neural networks learn complex non-linear features of images, and the architectural differences between systems suggest that these non-linear functions should take different forms. Nonetheless, these ten CNNs which were trained on the same data set seem to have learned to extract similar properties from images. In essence, each CNN's training algorithm hill-climbs in a very different parameter space, yet converges on a similar solution. This suggests that for CNNs, the selection of the training set and strategy may be more important than the selection of the convolutional architecture.
Open Access
Three types of sensory gating: exploring interrelationships, individual differences, and implications
(Colorado State University. Libraries, 2010) Yadon, Carly Ann, author; Davies, Patricia L., advisor; Nerger, Janice L., advisor; Anderson, Charles W., committee member; Cleary, Anne M., committee member
The primary purpose of this dissertation was to determine how information is selectively processed in the brain through sensory gating mechanisms. Filtering, habituation, and orienting are three types of sensory gating that have never been investigated together in the same study. Although it has been well established that sensory gating is abnormal in many clinical groups, there remains a fundamental lack of understanding regarding the mechanisms of gating. For example, the functional significance of sensory gating, as well as how different types of sensory gating are related to basic brain processes and to each other, is poorly understood. Using an event-related potential (ERP) paradigm, I measured P50, N100, and P200 filtering, habituation, and orienting and administered a sequence of neuropsychological measures of attention to forty-two healthy adults. I found that filtering, orienting, and habituation and the three ERP components had different patterns of results, suggesting that the three paradigms measured distinct types of sensory gating and that gating is a multistage process. For all three types of sensory gating, higher-level attention tasks tended to predict gating responses better than lower-level attention tasks. This dissertation demonstrated that sensory gating has functional importance and these three gating paradigms seem to reflect different types of gating that should be explored in their own right.
Open Access
Using locally observed swarm behaviors to infer global features of harsh environments
(Colorado State University. Libraries, 2021) Emmons, Megan R., author; Maciejewski, Anthony A., advisor; Chong, Edwin K. P., advisor; Anderson, Charles W., committee member; Young, Peter M., committee member
Robots in a swarm are programmed with individual behaviors but then interactions with the environment and other robots produce more complex, emergent swarm behaviors. A partial differential equation (PDE) can be used to accurately quantify the distribution of robots throughout the environment at any given time if the robots have simple individual behaviors and there are a finite number of potential environments. A least mean square algorithm can then be used to compare a given observation of the swarm distribution to the potential models to accurately identify the environment being explored. This technique affirms that there is a correlation between the individual robot behaviors, robot distribution, and the environment being explored. For more complex behaviors and environments, there is no closed-form model for the emergent behavior but there is still a correlation which can be used to infer one property if the other two are known. A simple, single-layer neural network can replace the PDE and be trained to correlate local observations of the robot distribution to the environment being explored. The neural network approach allows for more sophisticated robot behaviors, more varied environments, and is robust to variations in environment type and number of robots. By replacing the neural network with a simulated human rescuer who uses only locally observed velocity information to navigate a disaster scenario, the impact of fundamental swarm properties can be systematically explored. Further, the baseline swarm resilience can be quantified. Collectively, this development lays a foundation for using minimalist swarms, where robots have simple motions and no communication, to achieve collective sensing which can be leveraged in a variety of applications where no other robotic solutions currently exist.
Open Access
Utilizing network features to detect erroneous inputs
(Colorado State University. Libraries, 2020) Gorbett, Matthew, author; Blanchard, Nathaniel, advisor; Anderson, Charles W., committee member; King, Emily, committee member
Neural networks are vulnerable to a wide range of erroneous inputs such as corrupted, out-of-distribution, misclassified, and adversarial examples. Previously, separate solutions have been proposed for each of these faulty data types; however, in this work I show that the collective set of erroneous inputs can be jointly identified with a single model. Specifically, I train a linear SVM classifier to detect these four types of erroneous data using the hidden and softmax feature vectors of pre-trained neural networks. Results indicate that these faulty data types generally exhibit linearly separable activation properties from correctly processed examples. I am able to identify erroneous inputs with an AUROC of 0.973 on CIFAR10, 0.957 on Tiny ImageNet, and 0.941 on ImageNet. I experimentally validate the findings across a diverse range of datasets, domains, and pre-trained models.