Mountain Scholar :: Browsing by Author "Anderson, Chuck, committee member"

Browsing by Author "Anderson, Chuck, committee member"

Now showing 1 - 20 of 29

Open Access
A spectral analysis of the Crab nebula and other sources with HAWC
(Colorado State University. Libraries, 2016) Gussert, Michael, author; Harton, John, advisor; Mostafa, Miguel, advisor; Toki, Walter, committee member; Anderson, Chuck, committee member; Gelfand, Martin P., committee member
The High Altitude Water Cherenkov observatory (HAWC) is an extensive air shower particle detection array designed to study cosmic gamma (γ) rays in the Very High Energy (VHE) regime (100 GeV to 100 TeV). One of the most thoroughly studied sources in this energy range is the Crab nebula, a pulsar wind nebula created by the aftermath of supernova 1054. The core of this analysis revolves around the determination of the differential flux spectrum of the Crab nebula using a process known as forward folding. Forward folding allows energy spectra to be fit without requiring a direct measurement of the primary energy of individual extensive air showers. The energy resolution of HAWC is very poor (on the order of 50% or more), and so this method is ideal for any spectral analysis carried out with HAWC data. The differential spectra are modeled as a power law with a normalization (Φ0), spectral index (γ), and a cutoff energy (Ec): dN/dE = Φ0(E/E0)γe−E/Ec . The normalization of the Crab nebula was found to be 1.03±0.091 0.083 stat ±0.19 sys)×10−12(TeV−1 cm−2 s −1 ) with an index of −2.54 ± 0.095 stat ± 0.27 sys and a cutoff of 91.0 ±174 59 stat with E0 =4.0 TeV. This method was also applied to 11 other sources, and the minimum detection significance required to constrain a spectrum was found to be between 10 and 14 σ.
Open Access
A value-function based method for incorporating ensemble forecasts in real-time optimal reservoir operations
(Colorado State University. Libraries, 2020) Peacock, Matthew E., author; Labadie, John W., advisor; Ramirez, Jorge, committee member; Anderson, Chuck, committee member; Johnson, Lynn, committee member
Increasing stress on water resource systems has led to a desire to seek methods of improving the performance of reservoir operations. Water managers face many challenges including changes in demand, variable hydrological input and new environmental pressures. These issues have led to an interest in using ensemble streamflow forecasts to improve the performance of a reservoir system. The currently available methods for using ensemble forecasts encounter difficulties as the resolution of the analysis increases in order to accurately model a real-world system. One of the difficulties is due to the "curse of dimensionality'' as computing time exponentially increases when the discretization of the state and action spaces becomes finer or when more state or action variables are considered. Another difficulty is the problem of delayed rewards. When the time step of the analysis becomes shorter than the travel time due to routing, rewards may not be realized in the same time step as the action which caused them. Current methods such as dynamic programming or scenario-tree based methods are not able to handle delayed rewards. This research presents a value function-based method which separates the problem into two subproblems: computing the state-value function in the no-forecast condition, and finding optimal sequences of decisions given the ensemble forecast with the state-value function providing information about the value at any state at the end of the forecast horizon. A continuous action deep reinforcement learning algorithm is used to overcome the problems of dimensionality and delayed rewards, and a particle swarm method is used to find optimal decisions during the forecast horizon. The method is applied to a case study in the Russian River basin and compared to an idealized operating rule. The results show that the reinforcement learning process is able to generate policies that skillfully operate the reservoir without forecasts. When forecasts are used, the method is able to produce non-dominated performance measures. When the water stress to the system is increased by removing a transbasin diversion, the method outperforms the idealized operations.
Open Access
Accurate prediction of protein function using GOstruct
(Colorado State University. Libraries, 2011) Sokolov, Artem, author; Ben-Hur, Asa, advisor; Anderson, Chuck, committee member; McConnell, Ross M., committee member; Wang, Haonan, committee member
With the growing number of sequenced genomes, automatic prediction of protein function is one of the central problems in computational biology. Traditional methods employ transfer of functional annotation on the basis of sequence or structural similarity and are unable to effectively deal with today's noisy high-throughput biological data. Most of the approaches based on machine learning, on the other hand, break the problem up into a collection of binary classification problems, effectively asking the question ''does this protein perform this particular function?''; such methods often produce a set of predictions that are inconsistent with each other. In this work, we present GOstruct, a structured-output framework that answers the question ''what function does this protein perform?'' in the context of hierarchical multilabel classification. We show that GOstruct is able to effectively deal with a large number of disparate data sources from multiple species. Our empirical results demonstrate that the framework achieves state-of-the-art accuracy in two of the recent challenges in automatic function prediction: Mousefunc and CAFA.
Open Access
Application of an interpretable prototypical-part network to subseasonal-to-seasonal climate prediction over North America
(Colorado State University. Libraries, 2024) Gordillo, Nicolas J., author; Barnes, Elizabeth, advisor; Schumacher, Russ, committee member; Anderson, Chuck, committee member
In recent years, the use of neural networks for weather and climate prediction has greatly increased. In order to explain the decision-making process of machine learning "black-box" models, most research has focused on the use of machine learning explainability methods (XAI). These methods attempt to explain the decision-making process of the black box networks after they have been trained. An alternative approach is to build neural network architectures that are inherently interpretable. That is, construct networks that can be understood by a human throughout the entire decision-making process, rather than explained post-hoc. Here, we apply such a neural network architecture, named ProtoLNet, in a subseasonal-to-seasonal climate prediction setting. ProtoLNet identifies predictive patterns in the training data that can be used as prototypes to classify the input, while also accounting for the absolute location of the prototype in the input field. In our application, we use data from the Community Earth System Model version 2 (CESM2) pre-industrial long control simulation and train ProtoLNet to identify prototypes in precipitation anomalies over the Indian and North Pacific Oceans to forecast 2-meter temperature anomalies across the western coast of North America on subseasonal-to-seasonal timescales. These identified CESM2 prototypes are then projected onto fifth-generation ECMWF Reanalysis (ERA5) data to predict temperature anomalies in the observations several weeks ahead. We compare the performance of ProtoLNet between using CESM2 and ERA5 data. We then demonstrate a novel approach for performing transfer learning between CESM2 and ERA5 data which allows us to identify skillful prototypes in the observations. We show that the predictions by ProtoLNet using both datasets have skill while also being interpretable, sensible, and useful for drawing conclusions about what the model has learned.
Open Access
Artificial neural networks for fuel consumption and emissions modeling in light duty vehicles
(Colorado State University. Libraries, 2019) Chenna, Shiva Tarun, author; Jathar, Shantanu, advisor; Bradley, Thomas, committee member; Anderson, Chuck, committee member
There is growing evidence that real world, on-road emissions from mobile sources exceed emissions determined during laboratory tests and that the air quality, climate, and human health impacts from mobile sources might be substantially different than initially thought. Hence, there is an immediate need to measure and model these exceedances if we are to better understand and mitigate the environmental impacts of mobile sources. In this work, we used a portable emissions monitoring system (PEMS) and artificial neural networks (ANNs) to measure and model on-road fuel consumption and tailpipe emissions from Tier-2 light-duty gasoline and diesel vehicle. Tests were performed on at least five separate days for each vehicle and each test included a cold start and operation over a hot phase. Routes were deliberately picked to mimic certain features (e.g., distance, time duration) of driving cycles used for emissions certification (e.g., FTP-75). Data were gathered for a total of 49 miles and 145 minutes for the gasoline vehicle and 52 miles and 165 minutes for the diesel vehicle. Fuel consumption and emissions data were calculated at 1 Hz using information gathered from the vehicle using the onboard diagnostics port and the PEMS measurements. Route-integrated tailpipe emissions did not exceed the Tier-2 emissions standard for CO, NOX, and non-methane organic gases (NMOG) for either vehicle but did exceed so for PM for the diesel vehicle. We trained ANN models on part of the data to predict fuel consumption and tailpipe emissions at 1 Hz for both vehicles and evaluated these models against the rest of the data. The ANN models performed best when the training iterations (or epochs) were set to larger than 25 and the number of neurons in the hidden layer was between 7 and 9, although we did not see any specific advantage in increasing the number of hidden layers beyond 1. The trained ANN model predicted the fuel consumption over test routes within 5.5% of the measured value for both gasoline and diesel vehicles. The ANN performance varied significantly with pollutant type for the two vehicles and we were able to develop satisfactory models only for unburned hydrocarbons (HC) and NOX for diesel vehicles. Over independent test routes, the trained ANN models predicted HC within 12.5% of the measured value for the gasoline vehicle and predicted NOX emissions within 3% of the measured values for the diesel vehicle. The ANN performed better than, and hence could be used in lieu of, multivariable regression models such as those used in mobile source emissions models (e.g., EMFAC). In an 'environmental-routing' case study performed over three origin-destination pairs, the ANNs were able to successfully pick routes that minimized fuel consumption. Our work demonstrates the use of artificial neural networks to model fuel consumption and tailpipe emissions from light-duty passenger vehicles, with applications ranging from environmental routing to emissions inventory modeling.
Open Access
Causal inference using observational data - case studies in climate science
(Colorado State University. Libraries, 2020) Samarasinghe, Savini M., author; Ebert-Uphoff, Imme, advisor; Anderson, Chuck, committee member; Chong, Edwin, committee member; Kirby, Michael, committee member
We are in an era where atmospheric science is data-rich in both observations (e.g., satellite/ sensor data) and model output. Our goal with causal discovery is to apply suitable data science approaches to climate data to make inferences about the cause-effect relationships between climate variables. In this research, we focus on using observational studies, an approach that does not rely on controlled experiments, to infer cause-effect. Due to reasons such as latent variables, these observational studies do not allow us to prove causal relationships. Nevertheless, they provide data-driven hypotheses of the interactions, which can enable us to get insights into the salient interactions as well as the timescales at which they occur. Even though there are many different causal inference frameworks and methods that rely on observational studies, these approaches have not found widespread use within the climate or Earth science communities. To date, the most commonly used observational approaches include lagged correlation/regression analysis, as well as the bivariate Granger causality approach. We can attribute this lack of popularity to two main reasons. First is the inherent difficulty of inferring cause-effect in climate. Complex processes in the climate interact with each other at varying time spans. These interactions can be nonlinear, the distributions of relevant climate variables can be non-Gaussian, and the processes can be chaotic. A researcher interested in these causal inference problems has to face many challenges varying from identifying suitable variables, data, preprocessing and inference methods, as well as setting up the inference problem in a physically meaningful way. Also, the limited exposure and accessibility to modern causal inference approaches is another reason for their limited use within the climate science community. In this dissertation, we present three case studies related to causal inference in climate science, namely, (1) causal relationships between the Arctic temperature and mid-latitude circulations, (2) relationships between the Madden Julian Oscillation (MJO) and the North Atlantic Oscillation (NAO) and (3) the causal relationships between atmospheric disturbances of different spatial scales (e.g., Planetary vs. Synoptic). We use methods based on probabilistic graphical models to infer cause-effect, specifically constraint-based structure learning methods, and graphical Granger methods. For each case study, we analyze and document the scientific thought process of setting up the problem, the challenges faced, and how we have dealt with the challenges. The challenges discussed include, but not limited to, method selection, variable representation, and data preparation. We also present a successful high-dimensional study of causal discovery in spectral space. The main objectives of this research are to make causal inference methods more accessible to a researcher/climate scientist who is at entry-level to spatiotemporal causality and to promote more modern causal inference methods to the climate science community. The case studies, covering a wide range of questions and challenges, are meant to act as a resourceful starting point to a researcher interested in tackling more general causal inference problems in climate.
Open Access
Differentiating associations between tasks and outcomes in the human brain
(Colorado State University. Libraries, 2022) Nelson, Lauren, author; Seger, Carol, advisor; Thomas, Michael, committee member; Anderson, Chuck, committee member; Tompkins, Sara Anne, committee member
In order to successfully achieve their goals in a noisy and changing environment, organisms must continually learn both Pavlovian (stimulus-outcome or S-O) and instrumental (action-outcome or A-O) associations. A wide range of brain regions are implicated in reinforcement learning and decision-making, including the basal ganglia, medial prefrontal cortex, the dorsolateral prefrontal cortex (dlPFC), and the anterior cingulate cortex (ACC). One possible explanation of disparate findings is that activation depends on the nature of the action or response under consideration. To investigate representations of task-reward associations, subjects switched between an emotional judgement task and a spatial judgement task, combined with either a high or low level of reward. A general linear model (GLM) compared activation for different combinations of task and reward. A cluster in the mid-prefrontal cortex was more active for right versus left response, whereas a cluster in the midbrain near the brainstem was more active for left responses. Performance of the spatial task was associated with activation in the ventral occipital cortex and ventral prefrontal cortex. Clusters in the posterior parietal cortex and lateral prefrontal cortex were more active during the emotion task. Receiving a large reward was accompanied by activation in primary somatosensory cortex and auditory cortex, while receiving a low reward appeared to recruit the anterior cingulate cortex. Comparing trials which yielded a reward versus trials with no reward revealed activation in the dorsal prefrontal cortex. A 2-way ANOVA examining independent contributions of response and reward found an effect of response in cuneus and pre-cuneus, an effect of reward in anterior insula and sensorimotor cortex, and an interaction in the post-central gyrus. A 2-way ANOVA of task and reward found a main effect of task in several clusters in the medial occipital cortex, a main effect of reward in somatosensory cortex and anterior insula, and an interaction in the ventral occipital and anterior prefrontal cortex.
Open Access
Efficient multidimensional uncertainty quantification of high speed circuits using advanced polynomial chaos approaches
(Colorado State University. Libraries, 2016) Ahadi Dolatsara, Majid, author; Roy, Sourajeet, advisor; Notaros, Branislav, committee member; Anderson, Chuck, committee member; Pezeshki, Ali, committee member
With the scaling of VLSI technology to sub-45 nm levels, uncertainty in the nanoscale manufacturing processes and operating conditions have been found to result in unpredictable circuit behavior at the chip, package, and board levels of modern integrated microsystems. Hence, modeling the forward propagation of uncertainty from the device-level parameters to the system-level response of high-speed circuits and systems forms a crucial requirement of modern computer-aided design (CAD) tools. This thesis presents novel approaches based on the generalized polynomial chaos (gPC) theory for the efficient multidimensional uncertainty quantification of general distributed and lumped high-speed circuit networks. The key feature of this work is the development of approaches which are more efficient and/or accurate comparing to recently suggested uncertainty quantification approaches in the literature. Main contributions of this thesis are development of two individual approaches for improvement of the conventional linear regression uncertainty quantification approach, and development of a sparse polynomial expansion of the stochastic response in an uncertain system. The validity of this work is established through multiple numerical examples.
Embargo
Energy-aware workload management for geographically distributed data centers
(Colorado State University. Libraries, 2023) Hogade, Ninad, author; Pasricha, Sudeep, advisor; Siegel, Howard Jay, committee member; Maciejewski, Anthony, committee member; Anderson, Chuck, committee member
Cloud service providers are distributing data centers globally to reduce operating costs while also improving the quality of service by using intelligent cloud management strategies. The development of time-of-use electricity pricing and renewable energy source models has provided the means to reduce high cloud operating costs through intelligent geographical workload distribution. However, neglecting essential considerations such as data center cooling power, interference effects from workload co-location in servers, net-metering, peak demand pricing of electricity, data transfer costs, and data center queueing delay has led to sub-optimal results in prior work because these factors have a significant impact on cloud operating costs, performance, and carbon emissions. This dissertation presents a series of critical research studies addressing the vital issues of energy efficiency, carbon emissions reductions, and operating cost optimization in geographically distributed data centers. It scrutinizes different approaches to workload management, considering the diverse, dynamic, and complex nature of these environments. Starting from an exploration of energy cost minimization through sophisticated workload management techniques, the research extends to integrate network awareness into the problem, acknowledging data transfer costs and queuing delays. These works employ mathematical and game theoretic optimization to find effective solutions. Subsequently, a comprehensive survey of state-of-the-art Machine Learning (ML) techniques utilized in cloud management is discussed. Then, the dissertation traverses into the realm of Deep Reinforcement Learning (DRL) based optimization for efficient management of cloud resources and workloads. Finally, the study culminates in a novel game-theoretic DRL method, incorporating non-cooperative game theory principles to optimize the distribution of AI workloads, considering energy costs, data transfer costs, and carbon footprints. The dissertation holds significant implications for sustainable and cost-effective cloud data center workload management.
Unknown
Increasing BCI usability in the home: assessing the user and caregiver perspectives
(Colorado State University. Libraries, 2017) Bruegger, Katie, author; Davies, Patti, advisor; Roll, Marla, advisor; Sample, Pat, committee member; Anderson, Chuck, committee member
Objective. Despite research indicating that brain-computer interface (BCI) technology can be an effective option for persons with motor disabilities, BCI is currently not being used by this population on a regular basis. The purpose of this research is to determine the current usability of the BCI system in the user's home from the perspective of BCI users with motor disabilities and their caregivers in order to influence the future direction of BCI advancement to improve the usability of BCI technology for this population. Method. Within this study, there were four separate phases. In Phase 1, using feedback from five participants with motor disabilities and three caregivers for persons with motor disabilities, a questionnaire was developed for both BCI users and caregivers to assess the experience of setting up and using the BCI system. In Phase 2, these questionnaires were administered to five participants with motor disabilities and their caregivers a total of three times after experiencing the set-up and use of a P300 BCI system in the user's home on three separate occasions. In Phase 3, the responses of the questionnaires were analyzed and common themes and patterns were used to develop a list of questions to guide a focus group discussion. Finally, in Phase 4 a focus group consisting of three BCI users and two caregivers was held to gather more in depth information about the experience of using and setting up the BCI system. Throughout these phases, both quantitative and qualitative methods were used to analyze data. Results: Quantitative data analysis of questionnaire responses yielded no significant results; however, a variety of patterns were identified and within these patterns multiple patterns were found to approach significance. Relationships that approached significance included the difference between user and caregiver ratings for ease of use over time (z= -1.730, p= 0.084) and the difference between user and caregiver ratings for the likelihood of using or advocating for the use of BCI in the home on a regular basis (z= -1.792, p=0.073). These findings showed that ease of use decreased across the course of the three visits for both caregivers and users and that caregivers were more likely to advocate for the use of BCI on a regular basis than BCI users. When asked what area of life participants wanted this version and future versions of BCI to help increase users' current participation, the most common response chosen by users was environmental aids to daily living (EADL) while the most common response for caregivers was verbal and written communication. Qualitative analysis of the focus group provided answers to the three research questions (What are current barriers preventing BCI from being used by persons with disabilities on a regular basis in the home?; What are the aspects of BCI that BCI user's and caregivers enjoy?; And what are BCI users' and their caregivers' desires for the future of BCI) and yielded three emergent themes: the cost benefit analysis of BCI use, comparison of BCI to existing technology, and BCI and its relationship to independence. Conclusions. Although at its current state, none of the participants believed that the P300 BCI system would be a valuable addition to their life, users and caregivers agreed that in cases where the need was great enough, the challenges of using the system would be outweighed by its benefits. With the implementation of developments that decrease the challenges involved in setting up and using the system or the implementation of developments that increase the utility of real life applications, future BCI systems will become more practical options for a larger population.
Unknown
Laplacian Eigenmaps for time series analysis
(Colorado State University. Libraries, 2020) Rosse, Patrick J., author; Kirby, Michael, advisor; Peterson, Chris, committee member; Adams, Henry, committee member; Anderson, Chuck, committee member
With "Big Data" becoming more available in our day-to-day lives, it becomes necessary to make meaning of it. We seek to understand the structure of high-dimensional data that we are unable to easily plot. What shape is it? What points are "related" to each other? The primary goal is to simplify our understanding of the data both numerically and visually. First introduced by M. Belkin, and P. Niyogi in 2002, Laplacian Eigenmaps (LE) is a non-linear dimensional reduction tool that relies on the basic assumption that the raw data lies in a low-dimensional manifold in a high-dimensional space. Once constructed, the graph Laplacian is used to compute a low-dimensional representation of the data set that optimally preserves local neighborhood information. In this thesis, we present a detailed analysis of the method, the optimization problem it solves, and we put it to work on various time series data sets. We show that we are able to extract neighborhood features from a collection of time series, which allows us to cluster specific time series based on noticeable signatures within the raw data.
Open Access
Large-scale automated protein function prediction
(Colorado State University. Libraries, 2016) Kahanda, Indika, author; Ben-Hur, Asa, advisor; Anderson, Chuck, committee member; Draper, Bruce, committee member; Zhou, Wen, committee member
Proteins are the workhorses of life, and identifying their functions is a very important biological problem. The function of a protein can be loosely defined as everything it performs or happens to it. The Gene Ontology (GO) is a structured vocabulary which captures protein function in a hierarchical manner and contains thousands of terms. Through various wet-lab experiments over the years scientists have been able to annotate a large number of proteins with GO categories which reflect their functionality. However, experimentally determining protein functions is a highly resource-intensive task, and a large fraction of proteins remain un-annotated. Recently a plethora automated methods have emerged and their reasonable success in computationally determining the functions of proteins using a variety of data sources – by sequence/structure similarity or using various biological network data, has led to establishing automated function prediction (AFP) as an important problem in bioinformatics. In a typical machine learning problem, cross-validation is the protocol of choice for evaluating the accuracy of a classifier. But, due to the process of accumulation of annotations over time, we identify the AFP as a combination of two sub-tasks: making predictions on annotated proteins and making predictions on previously unannotated proteins. In our first project, we analyze the performance of several protein function prediction methods in these two scenarios. Our results show that GOstruct, an AFP method that our lab has previously developed, and two other popular methods: binary SVMs and guilt by association, find it hard to achieve the same level of accuracy on these two tasks compared to the performance evaluated through cross-validation, and that predicting novel annotations for previously annotated proteins is a harder problem than predicting annotations for uncharacterized proteins. We develop GOstruct 2.0 by proposing improvements which allows the model to make use of information of a protein's current annotations to better handle the task of predicting novel annotations for previously annotated proteins. Experimental results on yeast and human data show that GOstruct 2.0 outperforms the original GOstruct, demonstrating the effectiveness of the proposed improvements. Although the biomedical literature is a very informative resource for identifying protein function, most AFP methods do not take advantage of the large amount of information contained in it. In our second project, we conduct the first ever comprehensive evaluation on the effectiveness of literature data for AFP. Specifically, we extract co-mentions of protein-GO term pairs and bag-of-words features from the literature and explore their effectiveness in predicting protein function. Our results show that literature features are very informative of protein function but with further room for improvement. In order to improve the quality of automatically extracted co-mentions, we formulate the classification of co-mentions as a supervised learning problem and propose a novel method based on graph kernels. Experimental results indicate the feasibility of using this co-mention classifier as a complementary method that aids the bio-curators who are responsible for maintaining databases such as Gene Ontology. This is the first study of the problem of protein-function relation extraction from biomedical text. The recently developed human phenotype ontology (HPO), which is very similar to GO, is a standardized vocabulary for describing the phenotype abnormalities associated with human diseases. At present, only a small fraction of human protein coding genes have HPO annotations. But, researchers believe that a large portion of currently unannotated genes are related to disease phenotypes. Therefore, it is important to predict gene-HPO term associations using accurate computational methods. In our third project, we introduce PHENOstruct, a computational method that directly predicts the set of HPO terms for a given gene. We compare PHENOstruct with several baseline methods and show that it outperforms them in every respect. Furthermore, we highlight a collection of informative data sources suitable for the problem of predicting gene-HPO associations, including large scale literature mining data.
Open Access
Late residual neural networks: an approach to combat the dead ReLU problem
(Colorado State University. Libraries, 2022) Ernst, Matthew Frederick, author; Whitley, Darrell, advisor; Anderson, Chuck, committee member; Buchanan, Norm, committee member
The rectified linear unit (ReLU) activation function has been a staple tool in deep learning to increase the performance of deep neural network architectures. However, the ReLU activation function has trade-offs with its performance, specifically the dead ReLU problem caused by vanishing gradients. In this thesis, we introduce "late residual connections" a type of residual neural network with connections from each hidden layer connected directly to the output layer of a network. These residual connections improve convergence for neural networks by allowing more gradient flow to the hidden layers of a network.
Open Access
Learned perception systems for self-driving vehicles
(Colorado State University. Libraries, 2022) Chaabane, Mohamed, author; Beveridge, Ross J., advisor; O'Hara, Stephen, committee member; Blanchard, Nathaniel, committee member; Anderson, Chuck, committee member; Rebecca, Atadero, committee member
Building self-driving vehicles is one of the most impactful technological challenges of modern artificial intelligence. Self-driving vehicles are widely anticipated to revolutionize the way people and freight move. In this dissertation, we present a collection of work that aims to improve the capability of the perception module, an essential module for safe and reliable autonomous driving. Specifically, it focuses on two perception topics: 1) Geo-localization (mapping) of spatially-compact static objects, and 2) Multi-target object detection and tracking of moving objects in the scene. Accurately estimating the position of static objects, such as traffic lights, from the moving camera of a self-driving car is a challenging problem. In this dissertation, we present a system that improves the localization of static objects by jointly optimizing the components of the system via learning. Our system is comprised of networks that perform: 1) 5DoF object pose estimation from a single image, 2) association of objects between pairs of frames, and 3) multi-object tracking to produce the final geo-localization of the static objects within the scene. We evaluate our approach using a publicly available data set, focusing on traffic lights due to data availability. For each component, we compare against contemporary alternatives and show significantly improved performance. We also show that the end-to-end system performance is further improved via joint training of the constituent models. Next, we propose an efficient joint detection and tracking model named DEFT, or "Detection Embeddings for Tracking." The proposed approach relies on an appearance-based object matching network jointly learned with an underlying object detection network. An LSTM is also added to capture motion constraints. DEFT has comparable accuracy and speed to the top methods on 2D online tracking leaderboards while having significant advantages in robustness when applied to more challenging tracking data. DEFT raises the bar on the nuScenes monocular 3D tracking challenge, more than doubling the performance of the previous top method (3.8x on AMOTA, 2.1x on MOTAR). We analyze the difference in performance between DEFT and the next best-published method on nuScenes and find that DEFT is more robust to occlusions and large inter-frame displacements, making it a superior choice for many use-cases. Third, we present an end-to-end model to solve the tasks of detection, tracking, and sequence modeling from raw sensor data, called Attention-based DEFT. Attention-based DEFT extends the original DEFT by adding an attentional encoder module that uses attention to compute tracklet embedding that 1) jointly reasons about the tracklet dependencies and interaction with other objects present in the scene and 2) captures the context and temporal information of the tracklet's past observations. The experimental results show that Attention-based DEFT performs favorably against or comparable to state-of-the-art trackers. Reasoning about the interactions between the actors in the scene allows Attention-based DEFT to boost the model tracking performance in heavily crowded and complex interactive scenes. We validate the sequence modeling effectiveness of the proposed approach by showing its superiority for velocity estimation task over other baseline methods on both simple and complex scenes. The experiments demonstrate the effectiveness of Attention-based DEFT for capturing spatio-temporal interaction of the crowd for velocity estimation task, which helps it to be more robust to handle complexities in densely crowded scenes. The experimental results show that all the joint models in this dissertation perform better than solving each problem independently.
Open Access
Leveraging expression and network data for protein function prediction
(Colorado State University. Libraries, 2012) Graim, Kiley, author; Ben-Hur, Asa, advisor; Anderson, Chuck, committee member; Achter, Jeff, committee member
Protein function prediction is one of the prominent problems in bioinformatics today. Protein annotation is slowly falling behind as more and more genomes are being sequenced. Experimental methods are expensive and time consuming, which leaves computational methods to fill the gap. While computational methods are still not accurate enough to be used without human supervision, this is the goal. The Gene Ontology (GO) is a collection of terms that are the standard for protein function annotations. Because of the structure of GO, protein function prediction is a hierarchical multi-label classification problem. The classification method used in this thesis is GOstruct, which performs structured predictions that take into account all GO terms. GOstruct has been shown to work well, but there are still improvements to be made. In this thesis, I work to improve predictions by building new kernels from the data that are used by GOstruct. To do this, I find key representations of the data that help define what kernels perform best on the variety of data types. I apply this methodology to function prediction in two model organisms, Saccharomyces cerevisiae and Mus musculus, and found better methods for interpreting the data.
Open Access
Mean variants on matrix manifolds
(Colorado State University. Libraries, 2012) Marks, Justin D., author; Peterson, Chris, advisor; Kirby, Michael, advisor; Bates, Dan, committee member; Anderson, Chuck, committee member
The geometrically elegant Stiefel and Grassmann manifolds have become organizational tools for data applications, such as illumination spaces for faces in digital photography. Modern data analysis involves increasingly large-scale data sets, both in terms of number of samples and number of features per sample. In circumstances such as when large-scale data has been mapped to a Stiefel or Grassmann manifold, the computation of mean representatives for clusters of points on these manifolds is a valuable tool. We derive three algorithms for determining mean representatives for a cluster of points on the Stiefel manifold and the Grassmann manifold. Two algorithms, the normal mean and the projection mean, follow the theme of the Karcher mean, relying upon inversely related maps that operate between the manifold and the tangent bundle. These maps are informed by the geometric definition of the tangent bundle and the normal bundle. From the cluster of points, each algorithm exploits these maps in a predictor/corrector loop until converging, with prescribed tolerance, to a fixed point. The fixed point acts as the normal mean representative, or projection mean representative, respectively, of the cluster. This method shares its principal structural characteristics with the Karcher mean, but utilizes a distinct pair of inversely related maps. The third algorithm, called the flag mean, operates in a context comparable to a generalized Grassmannian. It produces a mean subspace of arbitrary dimension. We provide applications and discuss generalizations of these means to other manifolds.
Open Access
Microgrid optimization, modelling and control
(Colorado State University. Libraries, 2014) Han, Yi, author; Yount, Peter M., advisor; Chong, Edwin K. P., committee member; Pezeshki, Ali, committee member; Anderson, Chuck, committee member
To view the abstract, please see the full text of the document.
Open Access
Midlatitude prediction skill following QBO-MJO activity on subseasonal to seasonal timescales
(Colorado State University. Libraries, 2019) Mayer, Kirsten J., author; Barnes, Elizabeth A., advisor; Maloney, Eric, committee member; Anderson, Chuck, committee member
The Madden-Julian Oscillation (MJO) is known to force extratropical weather days-to-weeks following an MJO event through excitation of Rossby waves, also known as tropical-extratropical teleconnections. Prior research has demonstrated that this tropically forced midlatitude response can lead to increased prediction skill on subseasonal to seasonal (S2S) timescales. Furthermore, the Quasi-Biennial Oscillation (QBO) has been shown to possibly alter these teleconnections through modulation of the MJO itself and the atmospheric basic state upon which the Rossby waves propagate. This implies that the MJO-QBO relationship may affect midlatitude circulation prediction skill on S2S timescales. In this study, we quantify midlatitude circulation sensitivity and prediction skill following active MJOs and QBOs across the Northern Hemisphere on S2S timescales through an examination of the 500 hPa geopotential height field. First, a comparison of the spatial distribution of Northern Hemisphere sensitivity to the MJO during different QBO phases is performed for ERA-Interim reanalysis as well as ECMWF and NCEP hindcasts. Secondly, differences in prediction skill in ECMWF and NCEP hindcasts are quantified following MJO-QBO activity. We find that regions across the Pacific, North America and the Atlantic exhibit increased prediction skill following MJO-QBO activity, but these regions are not always collocated with the locations most sensitive to the MJO under a particular QBO state. Both hindcast systems demonstrate enhanced prediction skill 7-14 days following active MJO events during strong QBO periods compared to MJO events during neutral QBO periods.
Open Access
Neural correlates of executed and imagined joystick directional movements: a functional near-infrared spectroscopy study
(Colorado State University. Libraries, 2019) Mathison, Matthew A., author; Rojas, Donald C., advisor; Davalos, Deana, committee member; Anderson, Chuck, committee member
Motor-based brain computer interfaces (BCIs) attempt to restore and/or enhance motor functioning by measuring brain signals and converting them to computerized output. Functional near-infrared spectroscopy (fNIRS) is a non-invasive brain imaging modality that is resistant to both noise and motion-related artifacts. For this reason, fNIRS offers potential as an imaging method for use in a BCI. Currently, there is a paucity of literature on fNIRS as a sole BCI imaging method. Of the extant literature, studies were limited by low-density optode layouts and/or task designs which did not represent the motor goal. The present study was designed to enhance our understanding of the capabilities of fNIRS by utilizing a high-density optode array and an experimental task that closely mirrored the motor goal. 28 participants completed a series of executed and imagined joystick movements in four directions (forward, back, right, and left). Results indicated significant differences in inferred cortical activation during executed movements compared to baseline, executed movements compared to imagined movements, and imagined movements compared to baseline. No significant differences were observed for comparisons between individual movement directions. Results support the possibility that fNIRS may not be capable of distinguishing between changes in brain activity associated with joystick movement directions. Future research could enhance classification accuracy by implementing a machine learning algorithm or by pairing fNIRS with electroencephalography.
Open Access
Neuralator 5000: exploring and enhancing the BOLD5000 fMRI dataset to improve the robustness of artificial neural networks
(Colorado State University. Libraries, 2023) Pickard, William Augustus, author; Blanchard, Nathaniel, advisor; Anderson, Chuck, committee member; Thomas, Michael, committee member
Artificial neural networks (ANNs) originally drew their inspiration from biological constructs. Despite the rapid development of ANNs and their seeming divergence from their biological roots, research using representational similarity analysis (RSA) shows a connection between the internal representations of artificial and biological neural networks. To further investigate this connection, human subject functional magnetic resonance imaging (fMRI) studies using stimuli drawn from common ANN training datasets are being compiled. One such dataset is the BOLD5000, which is composed of fMRI data from four subjects who were presented with stimuli selected from the ImageNet, Common Objects in Context (COCO), and Scene UNderstanding (SUN) datasets. An important area where this data can be fruitful is in improving ANN model robustness. This work seeks to enhance the BOLD5000 dataset and make it more accessible for future ANN research by re-segmenting the data from the second release of the BOLD5000 into new ROIs using the vcAtlas and visfAtlas visual cortex atlases, generating representational dissimilarity matrices (RDMs) for all ROIs, and providing a new, biologically-inspired set of supercategory labels specific to the ImageNet dataset. To demonstrate the utility of these new BOLD5000 derivatives, I compare human fMRI data to RDMs derived from the activations of four prominent vision ANNs: AlexNet, ResNet-50, MobileNetV2, and EfficientNet B0. The results of this analysis show that the old, less-advanced AlexNet has a higher neuro-similarity than the much more recent, and technically better-performing models. These results are further confirmed through the use of Fiedler vector analysis on the RDMs, which shows a reduction in the separability of the internal representations of the biologically inspired supercategories.